Compare commits

..
Author SHA1 Message Date
01fc518b29 Python: bump package versions for 1.8.0 release (#6351)
- Released cohort (core, openai, foundry, root): 1.7.0 -> 1.8.0
- agent-framework-github-copilot: promote to RC (1.0.0rc1)
- agent-framework-orchestrations: rc2 -> rc3 (bug fix)
- Beta/alpha packages with changes: a2a, anthropic, azurefunctions, bedrock,
  foundry-hosting, mistral bumped to new date stamp (260604)
- Inter-package dependency bounds updated for changed packages
- CHANGELOG.md and PACKAGE_STATUS.md updated

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-04 23:03:24 +00:00
f3c3efed43 Python: Add GitHub Copilot integration tests to CI workflows (#6346)
Add a dedicated integration test job for the github_copilot package to both
python-integration-tests.yml and python-merge-tests.yml.

The job:
- Runs 6 integration tests marked with @pytest.mark.integration
- Uses COPILOT_GITHUB_TOKEN secret from the integration environment
- Follows the same pattern as other provider integration jobs
- Includes path filtering in merge-tests (github_copilot package + core changes)
- Added to needs lists in report and check jobs

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-04 22:06:26 +00:00
bbccb7c28c .NET: Bump ModelContextProtocol from 1.1.0 to 1.2.0 (#3956) (#6239)
Co-authored-by: Neeraj Karamchandani <neerajkaramchandani@mac.mynetworksettings.com>
2026-06-04 21:51:15 +01:00
Tao ChenandGitHub dbc312a78a Python: Fix toolbox consent flow in hosted agent (#6249)
* Fix toolbox consent flow in hosted agent

* Resolve conflict

* Make unused tool as comment

* Fix tests
2026-06-04 20:28:59 +00:00
bb9ed63a34 .NET: Restructure skill script schemas XML and remove resources from body (#6343)
* Restore UTF-8 BOMs and fix BuildScriptSchemasBlock doc comment

- Restore UTF-8 BOM on all changed files to match repo convention
- Fix XML doc: <schema name=...> -> <schema script=...> to match emitted output

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Address PR review comments: fix doc remarks and rename tests

- Update script doc remarks to clarify only parameter schemas are included
- Fix grammar: 'arguments format' -> 'argument format'
- Rename misleading test methods to match actual assertions
- Clarify comment about removed wrapper element

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: SergeyMenshykh <SergeMenshikh@outlook.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-04 21:15:29 +01:00
6b94315161 Python: Add timeout parameter to FoundryAgent to fix ConnectTimeout on multi-turn conversations (#6263)
* Python: fix ConnectTimeout on multi-turn FoundryAgent conversations (#6241)

Expose a `timeout` parameter on `RawFoundryAgentChatClient`,
`_FoundryAgentChatClient`, `RawFoundryAgent`, `FoundryAgent`, and
`RawOpenAIChatClient` so callers can override the HTTP timeout used by
the underlying AsyncOpenAI client.

Root cause: `RawFoundryAgentChatClient.__init__` called
`project_client.get_openai_client()` without configuring any timeout,
inheriting the OpenAI SDK default of `httpx.Timeout(connect=5.0)`.
When connections are recycled between turns under load, the 5 s connect
timeout fires and surfaces as `openai.APITimeoutError`.

Fix:
- `load_openai_service_settings` (`_shared.py`): accept `timeout` and
  include it in `client_args` for all three `AsyncOpenAI`/
  `AsyncAzureOpenAI` construction paths.
- `RawOpenAIChatClient.__init__` (`_chat_client.py`): accept `timeout`
  and forward to `load_openai_service_settings`.
- `RawFoundryAgentChatClient.__init__` (`_agent.py`): accept `timeout`
  and set `openai_client.timeout = timeout` on the client returned by
  `get_openai_client()` before passing it to the base class.
- `_FoundryAgentChatClient`, `RawFoundryAgent`, `FoundryAgent`: accept
  and propagate `timeout` through the construction chain.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Add timeout parameter to FoundryAgent and RawOpenAIChatClient

Expose a timeout parameter on RawFoundryAgentChatClient,
_FoundryAgentChatClient, RawFoundryAgent, FoundryAgent, and
RawOpenAIChatClient. When provided, the value is applied to the
underlying AsyncOpenAI client so that connect timeouts under load
or after connection recycling can be tuned by callers.

Previously, get_openai_client() was called without any timeout
override, so the SDK default of httpx.Timeout(connect=5.0) was
inherited and could fire on multi-turn conversations where the
underlying connection is recycled between turns.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Python: Add `timeout` parameter to `FoundryAgent` to fix `ConnectTimeout` on multi-turn conversations

Fixes #6241

* fix(foundry): use with_options to avoid mutating shared OpenAI client timeout (#6241)

Replace direct assignment  with
 in
RawFoundryAgentChatClient.__init__.

The Azure AI Projects SDK caches and returns a shared AsyncOpenAI client
per AIProjectClient. Mutating its .timeout attribute leaked the override
to all other code paths sharing that client (other agents, user code).
with_options() returns a new client instance with the override applied,
leaving the original shared client untouched.

Update tests to assert with_options is called with the correct timeout
and that the original shared client's timeout attribute is not mutated.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* test(foundry): assert with_options return value flows to instance.client (#6241)

The four timeout propagation tests verified that with_options was called
but did not confirm that the returned (timeout-configured) client was
actually stored on the instance. A silent discard of the return value
would have left the tests green while the timeout had no effect.

Each test now captures the constructed instance and asserts:
  assert <instance>.client is openai_client_mock.with_options.return_value

Affected tests:
- test_raw_foundry_agent_chat_client_init_applies_timeout_to_openai_client
- test_raw_foundry_agent_chat_client_init_applies_timeout_with_preview_enabled
- test_foundry_agent_chat_client_init_propagates_timeout
- test_foundry_agent_init_propagates_timeout_to_openai_client

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <copilot@github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-04 18:25:18 +00:00
Yufeng HeandGitHub bc0e65d716 fix: drop hosted MCP calls when reasoning is stripped (#6210) 2026-06-04 18:11:24 +00:00
4268080c20 Python: Fix spurious Magentic custom manager warning (#6261)
* Fix magentic manager warning

* Use typing_extensions.Sentinel for _MISSING sentinel value

Replace the bare object() sentinel with typing_extensions.Sentinel per
PEP 661 (now final). Sentinel provides a proper name and repr
('<_MISSING>') and is the idiomatic approach going forward.

Refs #4306

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* fix: correct Sentinel type annotation for max_stall_count param (#6261)

Use int | Sentinel for max_stall_count parameter type annotation instead
of int with cast(Any, _MISSING) to properly express that the parameter
can hold either an int or the _MISSING sentinel value. This fixes the
pyright reportUnnecessaryComparison errors caused by the types int and
Sentinel having no overlap.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Rename _MISSING sentinel to UNSET in orchestrations

The sentinel is user-visible as a default in public init signatures, so
use UNSET (no leading underscore) instead of the private _MISSING name.
Drop the now-unnecessary reportPrivateUsage ignores on the UNSET imports.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <copilot@github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-04 08:59:04 +00:00
fe08574a7c Python: [BREAKING] Upgrade github-copilot-sdk to v1.0.0 (stable) (#6292)
* Python: Upgrade github-copilot-sdk to v1.0.0 (stable)

Upgrade agent-framework-github-copilot from github-copilot-sdk 1.0.0b2 to the
stable 1.0.0 release, adapting to all breaking API changes.

Source changes (_agent.py):
- SubprocessConfig removed: use RuntimeConnection.for_stdio(path=...) +
  CopilotClient kwargs (connection, log_level, base_directory)
- Import paths: copilot.generated.session_events -> copilot.session_events
- Settings: copilot_home -> base_directory (env GITHUB_COPILOT_BASE_DIRECTORY)
- Default deny handler: PermissionDecisionUserNotAvailable() (from
  copilot.generated.rpc)

Test changes:
- Updated imports and client-construction assertions (kwargs-based)
- Permission handler tests use concrete decision types
  (PermissionDecisionApproveOnce, PermissionDecisionDeniedInteractivelyByUser)

Sample changes:
- Permission handlers use PermissionHandler.approve_all or sync
  approve_and_log pattern (v1.0.0 protocol v3 dispatch is incompatible
  with blocking input() in permission handlers)
- Function approval sample uses asyncio.to_thread for interactive prompts
- Simplified imports across all samples

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Address PR review: scope permission handlers, widen type, add test

- Shell sample: only approve kind='shell', deny others
- URL sample: only approve kind='url', deny others
- Use getattr() for kind-specific attributes to satisfy pyright
- Widen PermissionHandlerType to accept async handlers (matches SDK)
- Add test for _deny_all_permissions return value

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Fix validation script and strengthen test assertion

- Update scripts/sample_validation/create_dynamic_workflow_executor.py to
  use copilot.session_events imports and PermissionHandler.approve_all
- Assert isinstance(result, PermissionDecisionUserNotAvailable) instead of
  stringly-typed kind check

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Add integration tests for GitHubCopilotAgent

Add 6 integration tests mirroring .NET coverage:
- Basic non-streaming response
- Streaming response
- Function tool invocation
- Session context (multi-turn)
- Session resume by ID
- Shell command execution

Tests require COPILOT_GITHUB_TOKEN env var (skipped otherwise).
Each test cleans up its Copilot session via delete_session.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-04 08:42:35 +00:00
f970a699d8 Python: Fix compaction message-id collisions and tool-loop summary persistence (#6299)
* Fix compaction message-id collisions and tool-loop summary persistence

Fixes two bugs in the compaction strategies:

- #5237: incremental group annotation assigned message ids by position
  within the re-annotated slice, so moving the re-annotation start back to
  a previous group start restarted ids at 0 and produced collisions
  (e.g. a user message reusing an assistant message's id), merging groups
  and causing tool-result compaction to wrongly exclude messages.
  group_messages/_ensure_message_ids now take an id_offset and guard
  against existing-id collisions; annotate_message_groups threads the
  slice start index through as the offset.

- #4991: the function-invocation loop copied the message list each
  iteration, so summaries inserted by compaction landed in a throwaway
  copy and were lost across tool-loop iterations (only the persistent
  excluded flags survived). _prepare_messages_for_model_call now compacts
  the list in place when messages is a list, so inserted summaries persist.

Adds regression tests (incremental id uniqueness, existing-id collision
avoidance, idempotency, and tool-loop summary persistence including
streaming and conversation-id modes).

Also adds a summarization.py sample demonstrating SummarizationStrategy
directly with a real client, and reworks advanced.py with tool-call
groups and a real summarizer.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Guard incremental message-id assignment against prefix-id collisions

Addresses PR review on #5237: _ensure_message_ids only guarded against
collisions within the re-annotated slice. A preexisting (e.g. user-supplied)
id in the preserved prefix could still be reassigned in the suffix when the
id was numerically out of position, merging groups across the re-annotation
boundary again.

group_messages/_ensure_message_ids now accept reserved_ids, and
annotate_message_groups passes the preserved prefix's ids so auto-assigned
suffix ids never collide across the full list. Adds a regression test
reproducing the out-of-position prefix-id collision.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-04 08:37:59 +00:00
Yufeng HeandGitHub f29bae8fbc Python: run sync tools off the event loop (#5773)
* fix: run sync tools off event loop

* chore: silence harness tool marker type check
2026-06-04 04:42:08 +00:00
Peter IbekweandGitHub c3901a4ddd Fix Observability/WorkflowAsAnAgent sampl (#6316) 2026-06-03 23:52:50 +00:00
Evan MattsonandGitHub ba617fc3b5 Don't count dependabot prs as part of the limit (#6317) 2026-06-04 08:31:36 +09:00
afa7834e2e Updating dotnet package versions for 1.9 release (#6314)
Co-authored-by: Ben Thomas <25218250+alliscode@users.noreply.github.com>
2026-06-03 20:03:21 +00:00
c6951c21f6 Python: Add MCP-based skills discovery (McpSkillsSource) (#6169)
* Add MCP-based skills discovery (McpSkill, McpSkillsSource, McpSkillResource)

Implement Agent Skills discovery over MCP following the SEP-2640 convention:
- McpSkillsSource: reads skill://index.json to discover skills served by an MCP server
- McpSkill: lazily fetches SKILL.md content via resources/read on demand
- McpSkillResource: wraps MCP resource results (text and binary)
- Path traversal protection in get_resource for defense in depth
- Samples for Foundry Toolbox and standalone MCP skills server
- Comprehensive unit tests (514 lines)

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Address PR review comments: rename to MCP* convention, fix error handling and samples

- Rename McpSkill/McpSkillResource/McpSkillsSource to MCPSkill/MCPSkillResource/MCPSkillsSource
- Add data-URI prefix stripping for blob resource decoding
- Let non-McpError exceptions propagate from get_resource()
- Fix contradictory test comment
- Use interactive input() in mcp_based_skill sample
- Remove misleading sample output block

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Restore debug logging for McpError in get_resource()

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Use AzureCliCredential in Foundry toolbox skills sample for consistency

Replace DefaultAzureCredential with AzureCliCredential to match the
credential convention used in all other samples.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Use MCPStreamableHTTPTool in MCP skills sample

Replace raw mcp library imports (ClientSession, streamable_http_client)
with the framework's MCPStreamableHTTPTool to keep MCP server connections
consistent regardless of whether skills are enabled.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Branch on McpError.error.code so only not-found errors return empty

Previously _try_read_index() and get_resource() swallowed every McpError
as 'no skills available', making auth failures, server crashes, and
connection drops indistinguishable from a server that simply has no
skills.

Now only two codes are treated as not-found:
- -32002 (MCP-spec Resource not found)
- -32601 (METHOD_NOT_FOUND — server lacks resources/read)

All other McpError codes and non-McpError exceptions propagate with a
warning log, surfacing real failures visibly.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Add tests for non-McpError and non-not-found error propagation in MCP skills

Cover the re-raise branch in MCPSkill.get_resource for plain
ConnectionError/TimeoutError, the generic McpError (code 0) propagation
on get_resource, and TimeoutError propagation in _try_read_index.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Revert "Use MCPStreamableHTTPTool in MCP skills sample"

This reverts commit f31ed0ded9.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Introduce MCP_SKILLS experimental feature for MCP skill classes

Add a separate MCP_SKILLS feature ID to ExperimentalFeature enum and
use it for MCPSkillResource, MCPSkill, and MCPSkillsSource, since their
promotion timeline is partly outside of our control.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-03 18:09:50 +00:00
westeyandGitHub a982428916 .NET: Bug fixes for AGUI hosting and workflows (#6311)
* Add mcp tool execution fix

* Apply IsolationKeyScopedAgentSessionStore to MapAGUI by default if not yet set and improve comments in samples

* Address PR comments

* Fix formatting
2026-06-03 17:45:58 +00:00
90a3e5de47 .NET: Add ILoggerFactory and IServiceProvider to HarnessAgent constructor (#6273)
* Add ILoggerFactory and IServiceProvider to HarnessAgent constructor

Add optional ILoggerFactory and IServiceProvider parameters to the
HarnessAgent constructor and AsHarnessAgent extension method, passing
them to all downstream components that accept them:

- FunctionInvokingChatClient (via UseFunctionInvocation)
- CompactionProvider
- AgentSkillsProvider
- ChatClientAgent (via BuildAIAgent)
- AIAgentBuilder.Build()

Closes #6103

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Improve tests to verify ILoggerFactory and IServiceProvider propagation

- Add test verifying ILoggerFactory.CreateLogger() is called by
  downstream components (CompactionProvider, AgentSkillsProvider)
- Add test verifying IServiceProvider is queried during pipeline build

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-03 09:09:39 +00:00
49a6e433a3 Python: progressive tool exposure via FunctionInvocationContext (#6233)
* Python: progressive tool exposure via FunctionInvocationContext

Add first-class progressive tool exposure to the Python core function-calling
loop. Tools can now add or remove real FunctionTool schemas at runtime via the
injected FunctionInvocationContext, taking effect on the next iteration of the
loop.

- FunctionInvocationContext gains a live `tools` list plus experimental
  `add_tools()` / `remove_tools()` helpers (feature: PROGRESSIVE_TOOLS).
- The function-calling loop establishes a run-local, normalized tools list and
  threads it into the context at both invocation paths so mutations propagate.
- Add a sample (dynamic_tool_exposure.py) and a tools samples README, including
  a note that CodeAct providers (Monty/Hyperlight) use their own provider-level
  tool management instead.

Supersedes #3877.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Validate non-negative input in dynamic_tool_exposure sample tools

Address review feedback: factorial and fibonacci now return an error
message for negative n instead of producing incorrect results.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Make add_tools atomic and surface swallowed function errors

Address review feedback on progressive tool exposure:

- add_tools now validates the full batch against a throwaway copy before
  committing, so a duplicate-name clash partway through a sequence leaves
  the live tool list unchanged (all-or-nothing).
- _auto_invoke_function now logs a warning (with traceback) when a tool
  raises, so contract errors such as a duplicate-name ValueError from
  add_tools are debuggable without enabling include_detailed_errors.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Avoid retaining tracebacks when logging swallowed function errors

Logging with exc_info=exc fed the exception traceback to the logging
machinery, whose frame references created reference cycles collected
lazily by the cyclic GC. On Windows that could drop a hyperlight
WasmSandbox on a non-owning thread ("unsendable, dropped on another
thread"), crashing the xdist worker. Log a pre-formatted message with
the exception repr instead, so no traceback object is retained.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* added missing decorator

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-03 09:01:07 +00:00
Peter IbekweandGitHub 6086a74302 Python: Promote agent-framework-declarative package to RC (#6256)
* Promote agent-framework-declarative package to RC

* Update missed package status file.
2026-06-02 19:30:05 +00:00
fa8cfb7567 Python: Fix FoundryAgent stripping model from PromptAgent requests (#5526)
* Fix FoundryAgent stripping model from PromptAgent requests

Move run_options.pop('model', None) inside the _uses_foundry_agent_session()
conditional so that model is only stripped for hosted agent sessions (where
the server manages the model) and preserved for PromptAgent requests that
require it in the Responses API call.

Fixes #5525

* test: add coverage for resp_* continuation preserving model

Adds test_raw_foundry_agent_chat_client_prepare_options_preserves_model_for_resp_continuation
to explicitly verify that HostedAgent v1 / v2-no-session paths (where conversation_id
starts with resp_) preserve model and previous_response_id without triggering the
hosted-session gate.

---------

Co-authored-by: Benke Qu <bequ@microsoft.com>
Co-authored-by: Evan Mattson <35585003+moonbox3@users.noreply.github.com>
2026-06-02 18:30:04 +00:00
6de4c24fdd .NET: Promote Workflows.Declarative packages to stable versions (#6254)
* Promote Workflows.Declarative packages to stable versions

* Address PR feedback: enable package validation on GA declarative packages

Both Workflows.Declarative and Workflows.Declarative.Mcp set IsReleased=true

but were disabling package validation, bypassing the repo's GA convention

(see dotnet/nuget/nuget-package.props which auto-enables validation when

IsReleased=true).

Re-enable validation by removing the local EnablePackageValidation=false

overrides and pointing PackageValidationBaselineVersion at 1.8.0-rc1 (the

latest published version of each package). This catches accidental breaking

changes between RC and the first GA. Future GAs should bump the baseline to

the previous GA version.

Verified locally: dotnet build -c Release on both projects runs

RunPackageValidation -> APICompat ran successfully without finding any

breaking changes.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Update statement for the baseline validation.

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-02 15:10:02 +00:00
Dineshsuriya DandGitHub a5f355e04a Python: Fix OTLP HTTP base-endpoint losing /v1/{signal} auto-append (#5913)
* Python: Fix OTLP HTTP base-endpoint losing /v1/{signal} auto-append

Per the OTel spec, OTEL_EXPORTER_OTLP_ENDPOINT is a *base* URL for HTTP —
the SDK auto-appends /v1/traces, /v1/metrics, /v1/logs when it reads the
env var directly. Signal-specific endpoint env vars are *full* URLs used
verbatim.

_get_exporters_from_env read the base endpoint and forwarded it as the
constructor ``endpoint=`` argument, which the SDK always treats as a full
signal URL. As a result, with OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4318
and HTTP protocol, the exporter sent to http://localhost:4318 instead of
http://localhost:4318/v1/traces (and likewise for metrics/logs).

Replicate the spec's auto-append here when falling back to the base
endpoint under HTTP. gRPC behavior is unchanged.

* Python: Fix mypy type errors in OTLP endpoint assignment

Pre-declare traces_endpoint, metrics_endpoint, logs_endpoint as
str | None before the if/else block. Mypy inferred str from the
if-branch f-string assignments and then rejected the str | None
expressions in the else-branch as incompatible.
2026-06-02 09:59:50 +00:00
0cf48923cd .NET: Add Hosted-ToolboxMcpSkills sample (#6175)
* .NET: Add Hosted-ToolboxMcpSkills sample

Adds a hosted Foundry Responses sample that discovers MCP-based skills from a Foundry Toolbox and makes them available to the agent via AgentSkillsProvider.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Align README and Program.cs default model to gpt-5

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Clarify MCP skills provider log to avoid implying eager discovery

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Drop redundant skills provider configured log

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Add Foundry Toolbox Skills tag to manifest

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Simplify BearerTokenHandler by deriving from HttpClientHandler

Removes the need for an explicit InnerHandler. Enables CheckCertificateRevocationList to satisfy CA5399.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-02 08:41:21 +00:00
cdc4809b8a ci: harden Python test coverage workflow (#5982)
Improve input handling and token management in the Python test coverage
workflows.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-02 07:43:08 +00:00
Hameed KunkanoorandGitHub 043208241a Python: Persist hosted MCP call/results as canonical mcp_call output (#6070)
* Persist hosted MCP call/results as canonical mcp_call output

- Preserve hosted MCP call/result pairs as canonical mcp_call output items

- Coalesce MCP call + result in non-streaming conversion path

- Keep call-id alignment for MCP tool call tracking and output mapping

- Update tests and package metadata

* Fix missing Mapping import in hosted responses adapter

* Fix pyright unknown type in MCP output stringification

* Fix typing for MCP output sequence iteration

* Improve MCP output robustness and avoid eager flattening

* Bump foundry_hosting to b7 and update responses dependency to b7

* Restore foundry_hosting package version to 1.0.0a260521

* Refactor hosted MCP output parsing
2026-06-02 07:30:36 +00:00
Yufeng HeandGitHub 05ebb966cf fix: skip orphan anthropic thinking signatures (#5784) 2026-06-02 00:48:42 +00:00
Evan MattsonandGitHub c83a944e85 Fix open pr count check (#6255) 2026-06-02 09:09:36 +09:00
Thota Sai KarthikandGitHub 5d98beddf5 Python: feat(bedrock): implement native structured output support via Converse API (#6052)
* feat(bedrock): add structured output support via Converse API (Fixes #5966)

* fix(bedrock): improve unsupported model exception handling and schema parsing

* refactor(bedrock): use generic traversal for strict schema enforcement

* address Copilot review comments on structured output

* refine bedrock structured output: guard additionalProperties, TypeError check, docs + test

* fix(bedrock): widen response_format to Mapping and add missing test coverage
2026-06-01 23:30:19 +00:00
e0d0ad16a0 Python: feat(evals): Foundry Adaptive Evals integration (rubric-generation) (#6101)
* Python: feat(evals): RubricScore type + EvalScoreResult.dimensions

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Python: feat(foundry-evals): RubricDimension + GeneratedEvaluatorRef + accept in evaluators=

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Python: feat(evals): parse rubric_scores from output items + assertion helpers

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Python: feat(evals): BaseAgent.as_eval_source / Workflow.as_eval_source

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Python: feat(foundry-evals): EvalGenerationSource + generate_rubric helper

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Python: feat(foundry-evals): YAML config loader + sample

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Python: fix(evals): address PR review feedback

Addresses 4 Copilot review comments on PR #6101:

1. assert_dimension_score_at_least: drop the (not evaluator or found_any) guard so require_applicable=True correctly raises when the named evaluator produces no entries for the dimension. Adds TestRubricAssertions covering the regression.

2. GeneratedEvaluatorRef docstring: reword to describe actual behaviour (pinning recommended, not required) so it matches the dataclass default and FoundryEvals warning path.

3. _poll_generation_job: switch from asyncio.get_event_loop() to get_running_loop() and bound the per-iteration sleep by remaining time, matching _poll_eval_run.

4. generate_rubric: type category as Literal['quality','safety'] and validate at the entry point with a ValueError; drop the silent 'invalid -> quality' rewrite in _generation_job_to_ref. Adds a regression test.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Python: feat(foundry-evals): hosted-agent-aware rubric generation

* Auto-detect hosted Foundry agents in agent_as_eval_source: when the
  agent's chat_client exposes a string agent_name (the convention used
  by RawFoundryAgentChatClient for PromptAgents/HostedAgents), emit a
  type='agent' EvalGenerationSource so the service fetches instructions
  and tools from the agent registry instead of relying on the local
  wrapper (which holds neither for hosted agents).
* Add hosted_agent_version kwarg and a new agent_version field on
  EvalGenerationSource so PromptAgent runs can pin to a specific hosted
  version for reproducible rubric generation.
* Add force_prompt_source escape hatch to bypass auto-detection and
  always emit a rendered prompt dossier - useful when the local wrapper
  carries overrides the service-side agent doesnt see.
* Fix _to_sdk_source for dataset sources: SDK ctor takes name=/version=,
  not dataset_name=/dataset_version=. The mismatch would raise TypeError
  against the real azure-ai-projects 2.3.0a* SDK; only unmocked
  integration paths were affected.

Tests cover: auto-detection happy path, versionless hosted agent,
explicit hosted_agent_version forwarding, force_prompt_source override,
non-string chat_client attrs (MagicMock test doubles) not mis-detected,
agent_version forwarded through _to_sdk_source, and the corrected
dataset SDK kwarg names.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* fix(foundry-evals): accept canonical dimension_scores key per docs

The published Foundry rubric-evaluator output (Microsoft Learn 'Rubric evaluators' reference) places per-dimension breakdowns under properties.dimension_scores, not properties.rubric_scores. The parser now tries dimension_scores first and falls back to rubric_scores for preview-build compatibility, and tolerates non-list payloads (e.g. MagicMock auto-attrs) by trying the next candidate when parsing yields zero entries.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* feat(foundry-evals): add manual create_rubric_evaluator

Adds FoundryEvals.create_rubric_evaluator as the agent-framework surface over project_client.beta.evaluators.create_version. This is the manual counterpart to generate_rubric: callers supply RubricDimension instances (authored locally, ported from another framework, or hand-tuned) and we POST a RubricBasedEvaluatorDefinition. The service auto-attaches the non-editable residual dimension (general_quality for quality, general_policy_compliance for safety).

Per the Microsoft Learn 'Rubric evaluators' reference, the auto-generation path (create_generation_job) is primarily a portal/UI feature; external SDK clients with rich local agent context are better served by manual create_version. This keeps generate_rubric for users who want to round-trip through a Foundry-registered agent.

Validation up front: weight must be in [1,10], ids unique, descriptions non-empty, pass_threshold in [0,1]. The returned GeneratedEvaluatorRef is identical in shape to one obtained from generate_rubric, so downstream evaluators= lists work unchanged.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* samples(foundry-evals): manual rubric sample + namespace re-exports

Adds evaluate_with_manual_rubric_sample.py demonstrating the end-to-end dev scenario for FoundryEvals.create_rubric_evaluator: hand-author a list of RubricDimension, register via create_rubric_evaluator, then use the pinned GeneratedEvaluatorRef alongside built-in evaluators in an agent regression run.

Also re-exports RubricDimension, GeneratedEvaluatorRef, build_sources, and load_evals_config from agent_framework.foundry (both the lazy runtime shim and the type stub) so the rubric samples can import everything from a single namespace; the auto-generate sample was previously broken because the shim was missing build_sources / load_evals_config.

Updates the foundry-evals README with a chooser entry for the two rubric paths.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* feat(foundry-evals): remove rubric creation flows; keep consumption only

Reframes agent-framework as a pure consumer of Foundry rubric evaluators: scoring against rubrics that already exist (authored in the Foundry portal or via the dedicated SDK / REST surface) instead of creating them from the SDK.

Removed creation surface area:

- FoundryEvals.generate_rubric (auto-generate path) and create_rubric_evaluator (manual path), plus all _GenerationSdkTypes / _ManualRubricSdkTypes / _to_sdk_dimensions / _coalesce_generation_sources / _to_sdk_source / _poll_generation_job / _generation_job_to_ref / _evaluator_version_to_ref / _get_beta_evaluators / _import_*_sdk_types helpers.

- EvalGenerationSource (the input source discriminator), RubricDimension (the input dimension type), agent_as_eval_source / workflow_as_eval_source / _detect_hosted_foundry_agent helpers, and the YAML-config loader (_evals_config.py with RubricGenerationSpec / RubricSourceSpec / parse_evals_config / load_evals_config / build_sources).

- BaseAgent.as_eval_source / Workflow.as_eval_source plus the _render_agent_dossier / _render_workflow_dossier helpers in core. These existed only to feed the now-removed generation pipeline.

- Samples evaluate_with_generated_rubric_sample.py, evaluate_with_manual_rubric_sample.py, and evaluators.yaml. Replaced with a short README section showing how to reference an existing rubric evaluator via GeneratedEvaluatorRef.

Kept (consumption surface):

- GeneratedEvaluatorRef, slimmed to (name, version, display_name). Still accepted alongside built-in evaluator strings in FoundryEvals(evaluators=[...]). Versionless refs still warn.

- RubricScore on EvalScoreResult.dimensions plus EvalResults.assert_dimension_score_at_least for per-dimension CI gates.

- _parse_dimension_entries / _extract_rubric_scores output parsing (both canonical dimension_scores and the legacy rubric_scores key).

Tests: 160/160 foundry unit tests and 71/71 core local-eval tests pass; pyright is clean across changed files. The pre-existing tests/core/test_telemetry.py::test_detect_hosted_fallback_import_error failure is unrelated and reproduces on the prior commit.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* samples(foundry-evals): add evaluate_with_rubric_sample

Adds a runnable end-to-end sample showing how to consume a pre-existing rubric evaluator created in Foundry: reference it with GeneratedEvaluatorRef(name, version), mix it with built-in evaluators in FoundryEvals, and gate CI with assert_dimension_score_at_least on a specific dimension.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* fix(foundry-evals): satisfy mypy on _fetch_output_items

mypy infers OutputItemListResponse.sample as dict[str, object] | None while pyright correctly infers the typed Sample model. Cast to Any so both type checkers accept the attribute access pattern, rename the local to avoid shadowing the inner-loop sample binding, and drop the now-stale pyright suppressions.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* docs(foundry-evals): drop unpublished rubric-evaluators learn.microsoft.com link

The Adaptive Evals authoring docs are not yet published on Microsoft Learn, so the link 404s. Keep the descriptive text without the broken hyperlink; we can re-add it once the docs ship.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* test(foundry-evals): hoist repeated local imports to module top

Per code review feedback (eavanvalkenburg): the test file repeated 'from agent_framework_foundry._foundry_evals import ...' inside 22 test bodies and 'from agent_framework_foundry import GeneratedEvaluatorRef' inside 8 more. Move all of them to the existing top-level imports; the symbols are the same across tests and the local imports were redundant.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Ben Thomas <25218250+alliscode@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-01 23:01:56 +00:00
f36096ce1a Python: Fix core observability unsafe serialization of function-call arguments containing dataclass/framework objects (#6026)
* fix: safely serialize function-call arguments in core observability

Apply make_json_safe() to content.arguments in _to_otel_part() before
building the otel message dict, so that dataclass/framework payloads
(e.g. workflow request_info events) do not cause a TypeError when
_capture_messages() calls json.dumps().

Lift make_json_safe() into agent_framework._serialization (no new
external deps — dataclasses/datetime only) so the core observability
path can use it without a dependency on the ag-ui adapter.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* fix(core): safely serialize workflow request_info payloads in observability (#5733)

- Add make_json_safe() helper to recursively convert non-serializable objects
- Use make_json_safe() in _to_otel_part() for function_call arguments
- Fix CustomPayload test class to use @dataclass (resolves B903 lint error)

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* fix(serialization): guard callability and normalize dict keys in make_json_safe (#5733)

- Use callable(getattr(obj, method, None)) instead of hasattr() so that
  non-callable attributes named model_dump/to_dict/dict do not raise
  TypeError at runtime.
- Wrap each call in try/except TypeError to handle callables with
  mandatory arguments gracefully.
- Convert dict keys to str() so that non-string keys (e.g. datetime,
  int) cannot cause json.dumps to raise TypeError.
- Add regression tests for both scenarios.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Address observability serialization review feedback

---------

Co-authored-by: Copilot <copilot@github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-01 21:41:52 +00:00
03e14ca187 .NET: Update hosted agents (#6243)
* Updating to latest Foundry hosting packages.

* Re-applying .gitignore.

* Adding empty line at end of .gitignore

---------

Co-authored-by: Ben Thomas <25218250+alliscode@users.noreply.github.com>
2026-06-01 21:27:29 +00:00
160 changed files with 7395 additions and 912 deletions
+28 -10
View File
@@ -8,6 +8,7 @@ function getPullRequest(context) {
return {
author: pullRequest.user.login,
authorType: pullRequest.user.type,
labels: pullRequest.labels?.map((label) => label.name).filter(Boolean) ?? [],
number: pullRequest.number,
};
@@ -49,6 +50,10 @@ function hasLabel(labels, labelName) {
return labels.some((label) => label.toLowerCase() === labelName.toLowerCase());
}
function isDependabotAuthor({ author, authorType }) {
return authorType === 'Bot' && author.toLowerCase() === 'dependabot[bot]';
}
function buildLimitMessage({ author, exemptLabelName, maxOpenPrs, openPrCount }) {
return [
`Thank you for your contribution, @${author}.`,
@@ -63,24 +68,37 @@ function buildLimitMessage({ author, exemptLabelName, maxOpenPrs, openPrCount })
}
async function getOpenPrCount({ github, owner, repo, author, pullRequestNumber }) {
const query = `repo:${owner}/${repo} is:pr is:open author:${author}`;
const response = await github.rest.search.issuesAndPullRequests({
q: query,
const openPullRequests = await github.paginate(github.rest.pulls.list, {
owner,
repo,
state: 'open',
per_page: 100,
});
const indexedPrNumbers = response.data.items.map((item) => item.number);
const currentPrIsIndexed = indexedPrNumbers.includes(pullRequestNumber);
if (currentPrIsIndexed || response.data.total_count >= 100) {
return response.data.total_count;
}
const authorOpenPullRequestNumbers = openPullRequests
.filter((pullRequest) => pullRequest.user?.login === author)
.map((pullRequest) => pullRequest.number);
const currentPrIsOpen = authorOpenPullRequestNumbers.includes(pullRequestNumber);
const existingOpenPrCount = currentPrIsOpen
? authorOpenPullRequestNumbers.length - 1
: authorOpenPullRequestNumbers.length;
return response.data.total_count + 1;
return existingOpenPrCount + 1;
}
async function enforcePrLimit({ github, context, core, exemptLabelName, maxOpenPrs, labelName }) {
const { owner, repo } = context.repo;
const { author, labels, number } = getPullRequest(context);
const { author, authorType, labels, number } = getPullRequest(context);
if (isDependabotAuthor({ author, authorType })) {
core.info(`Author ${author} is Dependabot; skipping open PR limit enforcement.`);
return {
author,
closed: false,
dependabotExempt: true,
openPrCount: null,
};
}
if (hasLabel(labels, exemptLabelName)) {
core.info(`PR #${number} has the ${exemptLabelName} label; skipping open PR limit enforcement.`);
+83 -28
View File
@@ -16,7 +16,7 @@ const { enforcePrLimit } = require('../scripts/pr_limit_moderation.js');
// Helpers
// ---------------------------------------------------------------------------
function createContext({ author = 'community-user', labels = [], number = 123 } = {}) {
function createContext({ author = 'community-user', authorType = 'User', labels = [], number = 123 } = {}) {
return {
repo: {
owner: 'microsoft',
@@ -28,6 +28,7 @@ function createContext({ author = 'community-user', labels = [], number = 123 }
labels: labels.map((name) => ({ name })),
user: {
login: author,
type: authorType,
},
},
},
@@ -44,23 +45,20 @@ function createCore() {
};
}
function createGithub({ totalCount, itemNumbers, labelExists = true }) {
function createGithub({
itemNumbers,
labelExists = true,
pullRequests = createPullRequestPage({ numbers: itemNumbers }),
}) {
const calls = [];
return {
calls,
async paginate(method, params) {
calls.push({ api: 'paginate', method, params });
return pullRequests;
},
rest: {
search: {
async issuesAndPullRequests(params) {
calls.push({ api: 'search.issuesAndPullRequests', params });
return {
data: {
total_count: totalCount,
items: itemNumbers.map((number) => ({ number })),
},
};
},
},
issues: {
async getLabel(params) {
calls.push({ api: 'issues.getLabel', params });
@@ -85,6 +83,10 @@ function createGithub({ totalCount, itemNumbers, labelExists = true }) {
},
},
pulls: {
async list(params) {
calls.push({ api: 'pulls.list', params });
return { data: pullRequests };
},
async update(params) {
calls.push({ api: 'pulls.update', params });
return { data: { state: params.state } };
@@ -94,6 +96,15 @@ function createGithub({ totalCount, itemNumbers, labelExists = true }) {
};
}
function createPullRequestPage({ author = 'community-user', numbers }) {
return numbers.map((number) => ({
number,
user: {
login: author,
},
}));
}
// ---------------------------------------------------------------------------
// PR limit enforcement
@@ -102,7 +113,6 @@ function createGithub({ totalCount, itemNumbers, labelExists = true }) {
describe('PR limit enforcement', () => {
it('does not close the PR when the author is at the open PR limit', async () => {
const github = createGithub({
totalCount: 10,
itemNumbers: [1, 2, 3, 4, 5, 6, 7, 8, 9, 123],
});
@@ -119,14 +129,13 @@ describe('PR limit enforcement', () => {
assert.equal(result.openPrCount, 10);
assert.deepEqual(
github.calls.map((call) => call.api),
['search.issuesAndPullRequests'],
['paginate'],
);
});
it('counts the new PR when search has not indexed it yet', async () => {
it('counts the new PR when the pull list includes it', async () => {
const github = createGithub({
totalCount: 10,
itemNumbers: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
itemNumbers: [123, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
});
const result = await enforcePrLimit({
@@ -143,7 +152,7 @@ describe('PR limit enforcement', () => {
assert.deepEqual(
github.calls.map((call) => call.api),
[
'search.issuesAndPullRequests',
'paginate',
'issues.getLabel',
'issues.addLabels',
'issues.createComment',
@@ -152,9 +161,31 @@ describe('PR limit enforcement', () => {
);
});
it('counts the current PR on top of existing open PRs', async () => {
const github = createGithub({
itemNumbers: [123, ...Array.from({ length: 24 }, (_, index) => index + 1)],
pullRequests: createPullRequestPage({
numbers: [123, ...Array.from({ length: 25 }, (_, index) => index + 1)],
}),
});
const result = await enforcePrLimit({
github,
context: createContext(),
core: createCore(),
exemptLabelName: 'pr-limit-exempt',
maxOpenPrs: 10,
labelName: 'too-many-prs',
});
assert.equal(result.closed, true);
assert.equal(result.openPrCount, 26);
const comment = github.calls.find((call) => call.api === 'issues.createComment').params.body;
assert.match(comment, /This PR would put you at 26 open pull requests/);
});
it('creates the label when it does not already exist', async () => {
const github = createGithub({
totalCount: 11,
itemNumbers: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 123],
labelExists: false,
});
@@ -172,7 +203,7 @@ describe('PR limit enforcement', () => {
assert.deepEqual(
github.calls.map((call) => call.api),
[
'search.issuesAndPullRequests',
'paginate',
'issues.getLabel',
'issues.createLabel',
'issues.addLabels',
@@ -188,7 +219,6 @@ describe('PR limit enforcement', () => {
it('tolerates a 422 race when creating the label', async () => {
const github = createGithub({
totalCount: 11,
itemNumbers: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 123],
labelExists: false,
});
@@ -212,7 +242,7 @@ describe('PR limit enforcement', () => {
assert.deepEqual(
github.calls.map((call) => call.api),
[
'search.issuesAndPullRequests',
'paginate',
'issues.getLabel',
'issues.createLabel',
'issues.addLabels',
@@ -224,8 +254,11 @@ describe('PR limit enforcement', () => {
it('uses a diplomatic close message with the configured limit', async () => {
const github = createGithub({
totalCount: 11,
itemNumbers: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 123],
pullRequests: createPullRequestPage({
author: 'octo-contributor',
numbers: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 123],
}),
});
await enforcePrLimit({
@@ -246,7 +279,6 @@ describe('PR limit enforcement', () => {
it('does not close an exempt PR when it is reopened', async () => {
const github = createGithub({
totalCount: 11,
itemNumbers: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 123],
});
@@ -265,10 +297,33 @@ describe('PR limit enforcement', () => {
assert.deepEqual(github.calls, []);
});
it('does not over-count when the current PR is not on the first search page', async () => {
it('does not close Dependabot PRs', async () => {
const github = createGithub({
totalCount: 101,
itemNumbers: Array.from({ length: 100 }, (_, index) => index + 1),
itemNumbers: [123, ...Array.from({ length: 25 }, (_, index) => index + 1)],
pullRequests: createPullRequestPage({
author: 'dependabot[bot]',
numbers: [123, ...Array.from({ length: 25 }, (_, index) => index + 1)],
}),
});
const result = await enforcePrLimit({
github,
context: createContext({ author: 'dependabot[bot]', authorType: 'Bot' }),
core: createCore(),
exemptLabelName: 'pr-limit-exempt',
maxOpenPrs: 10,
labelName: 'too-many-prs',
});
assert.equal(result.closed, false);
assert.equal(result.dependabotExempt, true);
assert.equal(result.openPrCount, null);
assert.deepEqual(github.calls, []);
});
it('counts the current PR when the author has more than one page of open PRs', async () => {
const github = createGithub({
itemNumbers: [123, ...Array.from({ length: 100 }, (_, index) => index + 1)],
});
const result = await enforcePrLimit({
+42 -1
View File
@@ -474,6 +474,45 @@ jobs:
path: ./python/pytest.xml
if-no-files-found: ignore
# GitHub Copilot integration tests
python-tests-github-copilot:
name: Python Integration Tests - GitHub Copilot
runs-on: ubuntu-latest
environment: integration
timeout-minutes: 60
env:
COPILOT_GITHUB_TOKEN: ${{ secrets.COPILOT_GITHUB_TOKEN }}
GITHUB_COPILOT_TIMEOUT: "120"
defaults:
run:
working-directory: python
steps:
- uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6
with:
ref: ${{ inputs.checkout-ref }}
persist-credentials: false
- name: Set up python and install the project
id: python-setup
uses: ./.github/actions/python-setup
with:
python-version: ${{ env.UV_PYTHON }}
os: ${{ runner.os }}
- name: Test with pytest (GitHub Copilot integration)
run: >
uv run pytest --import-mode=importlib
packages/github_copilot/tests
-m integration
--timeout=120 --session-timeout=900 --timeout_method thread
--retries 2 --retry-delay 5
--junitxml=pytest.xml
- name: Upload test results
if: always()
uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
with:
name: test-results-github-copilot
path: ./python/pytest.xml
if-no-files-found: ignore
# Integration test trend report (aggregates per-job JUnit XML results)
python-integration-test-report:
name: Integration Test Report
@@ -490,6 +529,7 @@ jobs:
python-tests-foundry,
python-tests-foundry-hosting,
python-tests-cosmos,
python-tests-github-copilot,
]
runs-on: ubuntu-latest
defaults:
@@ -553,7 +593,8 @@ jobs:
python-tests-functions,
python-tests-foundry,
python-tests-foundry-hosting,
python-tests-cosmos
python-tests-cosmos,
python-tests-github-copilot
]
steps:
- name: Fail workflow if tests failed
+57
View File
@@ -40,6 +40,7 @@ jobs:
foundryChanged: ${{ steps.filter.outputs.foundry }}
foundryHostingChanged: ${{ steps.filter.outputs.foundry_hosting }}
cosmosChanged: ${{ steps.filter.outputs.cosmos }}
githubCopilotChanged: ${{ steps.filter.outputs.github_copilot }}
steps:
- uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6
- uses: dorny/paths-filter@d1c1ffe0248fe513906c8e24db8ea791d46f8590 # v3
@@ -85,6 +86,8 @@ jobs:
- 'python/packages/foundry_hosting/**'
cosmos:
- 'python/packages/azure-cosmos/**'
github_copilot:
- 'python/packages/github_copilot/**'
# run only if 'python' files were changed
- name: python tests
if: steps.filter.outputs.python == 'true'
@@ -658,6 +661,58 @@ jobs:
path: ./python/pytest.xml
if-no-files-found: ignore
# GitHub Copilot integration tests
python-tests-github-copilot:
name: Python Tests - GitHub Copilot Integration
needs: paths-filter
if: >
github.event_name != 'pull_request' &&
needs.paths-filter.outputs.pythonChanges == 'true' &&
(github.event_name != 'merge_group' ||
needs.paths-filter.outputs.githubCopilotChanged == 'true' ||
needs.paths-filter.outputs.coreChanged == 'true')
runs-on: ubuntu-latest
environment: integration
timeout-minutes: 60
env:
COPILOT_GITHUB_TOKEN: ${{ secrets.COPILOT_GITHUB_TOKEN }}
GITHUB_COPILOT_TIMEOUT: "120"
defaults:
run:
working-directory: python
steps:
- uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6
- name: Set up python and install the project
id: python-setup
uses: ./.github/actions/python-setup
with:
python-version: ${{ env.UV_PYTHON }}
os: ${{ runner.os }}
- name: Test with pytest (GitHub Copilot integration)
run: >
uv run pytest --import-mode=importlib
packages/github_copilot/tests
-m integration
--timeout=120 --session-timeout=900 --timeout_method thread
--retries 2 --retry-delay 5
--junitxml=pytest.xml
- name: Surface failing tests
if: always()
uses: pmeier/pytest-results-action@20b595761ba9bf89e115e875f8bc863f913bc8ad # v0.7.2
with:
path: ./python/pytest.xml
summary: true
display-options: fEX
fail-on-empty: false
title: GitHub Copilot integration test results
- name: Upload test results
if: always()
uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
with:
name: test-results-github-copilot
path: ./python/pytest.xml
if-no-files-found: ignore
# Integration test trend report (aggregates per-job JUnit XML results)
python-integration-test-report:
name: Integration Test Report
@@ -674,6 +729,7 @@ jobs:
python-tests-foundry,
python-tests-foundry-hosting,
python-tests-cosmos,
python-tests-github-copilot,
]
runs-on: ubuntu-latest
defaults:
@@ -735,6 +791,7 @@ jobs:
python-tests-foundry,
python-tests-foundry-hosting,
python-tests-cosmos,
python-tests-github-copilot,
]
steps:
- name: Fail workflow if tests failed
@@ -8,6 +8,7 @@ on:
permissions:
contents: read
actions: read
pull-requests: write
jobs:
@@ -23,7 +24,7 @@ jobs:
- name: Download coverage report
uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8
with:
github-token: ${{ secrets.GH_ACTIONS_PR_WRITE }}
github-token: ${{ github.token }}
run-id: ${{ github.event.workflow_run.id }}
path: ./python
merge-multiple: true
@@ -38,9 +39,9 @@ jobs:
echo "PR number file 'pr_number' is missing or empty"
exit 1
fi
PR_NUMBER=$(head -1 pr_number | tr -dc '0-9')
if [ -z "$PR_NUMBER" ]; then
echo "PR number file 'pr_number' does not contain a valid PR number"
PR_NUMBER=$(cat pr_number)
if ! [[ "$PR_NUMBER" =~ ^[0-9]+$ ]]; then
echo "::error::PR number file contains invalid content"
exit 1
fi
echo "PR_NUMBER=$PR_NUMBER" >> "$GITHUB_ENV"
@@ -48,7 +49,7 @@ jobs:
id: coverageComment
uses: MishaKav/pytest-coverage-comment@26f986d2599c288bb62f623d29c2da98609e9cd4 # v1.6.0
with:
github-token: ${{ secrets.GH_ACTIONS_PR_WRITE }}
github-token: ${{ github.token }}
issue-number: ${{ env.PR_NUMBER }}
pytest-xml-coverage-path: python/python-coverage.xml
title: "Python Test Coverage Report"
+1
View File
@@ -248,3 +248,4 @@ dotnet/filtered-*.slnx
.omx/
**/issues/
.test_*
+17 -17
View File
@@ -1,17 +1,17 @@
# Support
## How to file issues and get help
This project uses GitHub Issues to track bugs and feature requests. Please search the existing
issues before filing new issues to avoid duplicates. For new issues, file your bug or
feature request as a new Issue.
For help and questions about using this project, please create a GitHub issue.
AI Support team will support Microsoft Agent Framework issues for customers under a **Unified support agreement when the issue arises from usage of Azure AI services** (Foundry Models, Foundry Agents etc.) in conjunction with the SDK. Conversely, if customer has any other / non unified support agreement and/or Agent Framework SDK is used in a way **not involving an Azure service**, it is treated as a purely open-source tool – Microsoft’s support organization will not handle it, and users should use GitHub or forums for assistance
For Copilot Studio SDK implementation issues, customers should use GitHub Issues for assistance, as outlined above. Conversely, for prerequisites managed within the Copilot Studio portal, customers can rely on the standard Microsoft Copilot Studio support channels.
## Microsoft Support Policy
Support for this **PROJECT or PRODUCT** is limited to the resources listed above.
# Support
## How to file issues and get help
This project uses GitHub Issues to track bugs and feature requests. Please search the existing
issues before filing new issues to avoid duplicates. For new issues, file your bug or
feature request as a new Issue.
For help and questions about using this project, please create a GitHub issue.
AI Support team will support Microsoft Agent Framework issues for customers under a **Unified support agreement when the issue arises from usage of Azure AI services** (Foundry Models, Foundry Agents etc.) in conjunction with the SDK. Conversely, if customer has any other / non unified support agreement and/or Agent Framework SDK is used in a way **not involving an Azure service**, it is treated as a purely open-source tool – Microsoft’s support organization will not handle it, and users should use GitHub or forums for assistance
For Copilot Studio SDK implementation issues, customers should use GitHub Issues for assistance, as outlined above. Conversely, for prerequisites managed within the Copilot Studio portal, customers can rely on the standard Microsoft Copilot Studio support channels.
## Microsoft Support Policy
Support for this **PROJECT or PRODUCT** is limited to the resources listed above.
+1 -1
View File
@@ -109,7 +109,7 @@
<PackageVersion Include="A2A" Version="1.0.0-preview2" />
<PackageVersion Include="A2A.AspNetCore" Version="1.0.0-preview2" />
<!-- MCP -->
<PackageVersion Include="ModelContextProtocol" Version="1.1.0" />
<PackageVersion Include="ModelContextProtocol" Version="1.2.0" />
<!-- Hyperlight -->
<PackageVersion Include="Hyperlight.HyperlightSandbox.Api" Version="0.4.0" />
<PackageVersion Include="Hyperlight.HyperlightSandbox.Guest.Python" Version="0.4.0" />
+3
View File
@@ -344,6 +344,9 @@
<Folder Name="/Samples/04-hosting/FoundryHostedAgents/responses/Hosted-Toolbox/">
<Project Path="samples/04-hosting/FoundryHostedAgents/responses/Hosted-Toolbox/HostedToolbox.csproj" />
</Folder>
<Folder Name="/Samples/04-hosting/FoundryHostedAgents/responses/Hosted-ToolboxMcpSkills/">
<Project Path="samples/04-hosting/FoundryHostedAgents/responses/Hosted-ToolboxMcpSkills/HostedToolboxMcpSkills.csproj" />
</Folder>
<Folder Name="/Samples/04-hosting/FoundryHostedAgents/responses/Hosted-AzureSearchRag/">
<Project Path="samples/04-hosting/FoundryHostedAgents/responses/Hosted-AzureSearchRag/HostedAzureSearchRag.csproj" />
</Folder>
+3 -3
View File
@@ -1,14 +1,14 @@
<Project>
<PropertyGroup>
<!-- Central version prefix - applies to all nuget packages. -->
<VersionPrefix>1.8.0</VersionPrefix>
<VersionPrefix>1.9.0</VersionPrefix>
<RCNumber>1</RCNumber>
<DateSuffix>260528</DateSuffix>
<DateSuffix>260603</DateSuffix>
<PackageVersion Condition="'$(IsReleaseCandidate)' == 'true'">$(VersionPrefix)-rc$(RCNumber)</PackageVersion>
<PackageVersion Condition="'$(IsReleaseCandidate)' != 'true' AND '$(VersionSuffix)' != ''">$(VersionPrefix)-$(VersionSuffix).$(DateSuffix).1</PackageVersion>
<PackageVersion Condition="'$(IsReleaseCandidate)' != 'true' AND '$(VersionSuffix)' == ''">$(VersionPrefix)-preview.$(DateSuffix).1</PackageVersion>
<PackageVersion Condition="'$(IsReleased)' == 'true'">$(VersionPrefix)</PackageVersion>
<GitTag>1.8.0</GitTag>
<GitTag>1.9.0</GitTag>
<Configurations>Debug;Release;Publish</Configurations>
<IsPackable>true</IsPackable>
@@ -10,6 +10,11 @@ WebApplicationBuilder builder = WebApplication.CreateBuilder(args);
builder.Services.AddHttpClient().AddLogging();
builder.Services.AddAGUI();
// WARNING: When adding session persistence (e.g., WithInMemorySessionStore), or running in production,
// make sure to also register a SessionIsolationKeyProvider to scope sessions by principal in multi-user
// deployments, e.g.:
// builder.Services.UseClaimsBasedSessionIsolation(new() { ClaimType = ClaimTypes.NameIdentifier });
WebApplication app = builder.Build();
string endpoint = builder.Configuration["AZURE_OPENAI_ENDPOINT"]
@@ -14,6 +14,7 @@
<ItemGroup>
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.Hosting.AGUI.AspNetCore\Microsoft.Agents.AI.Hosting.AGUI.AspNetCore.csproj" />
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.Hosting.AspNetCore\Microsoft.Agents.AI.Hosting.AspNetCore.csproj" />
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.OpenAI\Microsoft.Agents.AI.OpenAI.csproj" />
</ItemGroup>
@@ -16,6 +16,11 @@ builder.Services.ConfigureHttpJsonOptions(options =>
options.SerializerOptions.TypeInfoResolverChain.Add(SampleJsonSerializerContext.Default));
builder.Services.AddAGUI();
// WARNING: When adding session persistence (e.g., WithInMemorySessionStore), or running in production,
// make sure to also register a SessionIsolationKeyProvider to scope sessions by principal in multi-user
// deployments, e.g.:
// builder.Services.UseClaimsBasedSessionIsolation(new() { ClaimType = ClaimTypes.NameIdentifier });
WebApplication app = builder.Build();
string endpoint = builder.Configuration["AZURE_OPENAI_ENDPOINT"]
@@ -14,6 +14,7 @@
<ItemGroup>
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.Hosting.AGUI.AspNetCore\Microsoft.Agents.AI.Hosting.AGUI.AspNetCore.csproj" />
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.Hosting.AspNetCore\Microsoft.Agents.AI.Hosting.AspNetCore.csproj" />
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.OpenAI\Microsoft.Agents.AI.OpenAI.csproj" />
</ItemGroup>
@@ -10,6 +10,11 @@ WebApplicationBuilder builder = WebApplication.CreateBuilder(args);
builder.Services.AddHttpClient().AddLogging();
builder.Services.AddAGUI();
// WARNING: When adding session persistence (e.g., WithInMemorySessionStore), or running in production,
// make sure to also register a SessionIsolationKeyProvider to scope sessions by principal in multi-user
// deployments, e.g.:
// builder.Services.UseClaimsBasedSessionIsolation(new() { ClaimType = ClaimTypes.NameIdentifier });
WebApplication app = builder.Build();
string endpoint = builder.Configuration["AZURE_OPENAI_ENDPOINT"]
@@ -14,6 +14,7 @@
<ItemGroup>
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.Hosting.AGUI.AspNetCore\Microsoft.Agents.AI.Hosting.AGUI.AspNetCore.csproj" />
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.Hosting.AspNetCore\Microsoft.Agents.AI.Hosting.AspNetCore.csproj" />
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.OpenAI\Microsoft.Agents.AI.OpenAI.csproj" />
</ItemGroup>
@@ -27,6 +27,11 @@ builder.Services.ConfigureHttpJsonOptions(options =>
options.SerializerOptions.TypeInfoResolverChain.Add(ApprovalJsonContext.Default));
builder.Services.AddAGUI();
// WARNING: When adding session persistence (e.g., WithInMemorySessionStore), or running in production,
// make sure to also register a SessionIsolationKeyProvider to scope sessions by principal in multi-user
// deployments, e.g.:
// builder.Services.UseClaimsBasedSessionIsolation(new() { ClaimType = ClaimTypes.NameIdentifier });
WebApplication app = builder.Build();
app.UseHttpLogging();
@@ -14,6 +14,7 @@
<ItemGroup>
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.Hosting.AGUI.AspNetCore\Microsoft.Agents.AI.Hosting.AGUI.AspNetCore.csproj" />
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.Hosting.AspNetCore\Microsoft.Agents.AI.Hosting.AspNetCore.csproj" />
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.OpenAI\Microsoft.Agents.AI.OpenAI.csproj" />
</ItemGroup>
@@ -17,6 +17,11 @@ builder.Services.AddAGUI();
// Configure to listen on port 8888
builder.WebHost.UseUrls("http://localhost:8888");
// WARNING: When adding session persistence (e.g., WithInMemorySessionStore), or running in production,
// make sure to also register a SessionIsolationKeyProvider to scope sessions by principal in multi-user
// deployments, e.g.:
// builder.Services.UseClaimsBasedSessionIsolation(new() { ClaimType = ClaimTypes.NameIdentifier });
WebApplication app = builder.Build();
string endpoint = builder.Configuration["AZURE_OPENAI_ENDPOINT"]
@@ -14,6 +14,7 @@
<ItemGroup>
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.Hosting.AGUI.AspNetCore\Microsoft.Agents.AI.Hosting.AGUI.AspNetCore.csproj" />
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.Hosting.AspNetCore\Microsoft.Agents.AI.Hosting.AspNetCore.csproj" />
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.OpenAI\Microsoft.Agents.AI.OpenAI.csproj" />
</ItemGroup>
@@ -50,12 +50,16 @@ internal static partial class WorkflowHelper
/// <summary>
/// Executor that starts the concurrent processing by sending messages to the agents.
/// </summary>
private sealed partial class ConcurrentStartExecutor() : Executor("ConcurrentStartExecutor")
[SendsMessage(typeof(List<ChatMessage>))]
[SendsMessage(typeof(TurnToken))]
private sealed partial class ConcurrentStartExecutor()
: Executor("ConcurrentStartExecutor", declareCrossRunShareable: true), IResettableExecutor
{
[MessageHandler]
internal ValueTask RouteMessages(List<ChatMessage> messages, IWorkflowContext context, CancellationToken cancellationToken)
internal ValueTask RouteMessages(IEnumerable<ChatMessage> messages, IWorkflowContext context, CancellationToken cancellationToken)
{
return context.SendMessageAsync(messages, cancellationToken: cancellationToken);
List<ChatMessage> payload = messages as List<ChatMessage> ?? messages.ToList();
return context.SendMessageAsync(payload, cancellationToken: cancellationToken);
}
[MessageHandler]
@@ -63,13 +67,16 @@ internal static partial class WorkflowHelper
{
return context.SendMessageAsync(token, cancellationToken: cancellationToken);
}
public ValueTask ResetAsync() => default;
}
/// <summary>
/// Executor that aggregates the results from the concurrent agents.
/// </summary>
[YieldsOutput(typeof(List<ChatMessage>))]
private sealed partial class ConcurrentAggregationExecutor() : Executor<List<ChatMessage>>("ConcurrentAggregationExecutor")
[YieldsOutput(typeof(string))]
private sealed partial class ConcurrentAggregationExecutor() :
Executor<List<ChatMessage>>("ConcurrentAggregationExecutor"), IResettableExecutor
{
private readonly List<ChatMessage> _messages = [];
@@ -90,5 +97,11 @@ internal static partial class WorkflowHelper
await context.YieldOutputAsync(formattedMessages, cancellationToken);
}
}
public ValueTask ResetAsync()
{
this._messages.Clear();
return default;
}
}
}
@@ -1,10 +0,0 @@
{
"IsEncrypted": false,
"Values": {
"FUNCTIONS_WORKER_RUNTIME": "dotnet-isolated",
"AzureWebJobsStorage": "UseDevelopmentStorage=true",
"DURABLE_TASK_SCHEDULER_CONNECTION_STRING": "Endpoint=http://localhost:8080;TaskHub=default;Authentication=None",
"AZURE_OPENAI_ENDPOINT": "<AZURE_OPENAI_ENDPOINT>",
"AZURE_OPENAI_DEPLOYMENT_NAME": "<AZURE_OPENAI_DEPLOYMENT_NAME>"
}
}
@@ -1,10 +0,0 @@
{
"IsEncrypted": false,
"Values": {
"FUNCTIONS_WORKER_RUNTIME": "dotnet-isolated",
"AzureWebJobsStorage": "UseDevelopmentStorage=true",
"DURABLE_TASK_SCHEDULER_CONNECTION_STRING": "Endpoint=http://localhost:8080;TaskHub=default;Authentication=None",
"AZURE_OPENAI_ENDPOINT": "<AZURE_OPENAI_ENDPOINT>",
"AZURE_OPENAI_DEPLOYMENT_NAME": "<AZURE_OPENAI_DEPLOYMENT_NAME>"
}
}
@@ -1,10 +0,0 @@
{
"IsEncrypted": false,
"Values": {
"FUNCTIONS_WORKER_RUNTIME": "dotnet-isolated",
"AzureWebJobsStorage": "UseDevelopmentStorage=true",
"DURABLE_TASK_SCHEDULER_CONNECTION_STRING": "Endpoint=http://localhost:8080;TaskHub=default;Authentication=None",
"AZURE_OPENAI_ENDPOINT": "<AZURE_OPENAI_ENDPOINT>",
"AZURE_OPENAI_DEPLOYMENT_NAME": "<AZURE_OPENAI_DEPLOYMENT_NAME>"
}
}
@@ -1,10 +0,0 @@
{
"IsEncrypted": false,
"Values": {
"FUNCTIONS_WORKER_RUNTIME": "dotnet-isolated",
"AzureWebJobsStorage": "UseDevelopmentStorage=true",
"DURABLE_TASK_SCHEDULER_CONNECTION_STRING": "Endpoint=http://localhost:8080;TaskHub=default;Authentication=None",
"AZURE_OPENAI_ENDPOINT": "<AZURE_OPENAI_ENDPOINT>",
"AZURE_OPENAI_DEPLOYMENT_NAME": "<AZURE_OPENAI_DEPLOYMENT_NAME>"
}
}
@@ -1,10 +0,0 @@
{
"IsEncrypted": false,
"Values": {
"FUNCTIONS_WORKER_RUNTIME": "dotnet-isolated",
"AzureWebJobsStorage": "UseDevelopmentStorage=true",
"DURABLE_TASK_SCHEDULER_CONNECTION_STRING": "Endpoint=http://localhost:8080;TaskHub=default;Authentication=None",
"AZURE_OPENAI_ENDPOINT": "<AZURE_OPENAI_ENDPOINT>",
"AZURE_OPENAI_DEPLOYMENT_NAME": "<AZURE_OPENAI_DEPLOYMENT_NAME>"
}
}
@@ -1,10 +0,0 @@
{
"IsEncrypted": false,
"Values": {
"FUNCTIONS_WORKER_RUNTIME": "dotnet-isolated",
"AzureWebJobsStorage": "UseDevelopmentStorage=true",
"DURABLE_TASK_SCHEDULER_CONNECTION_STRING": "Endpoint=http://localhost:8080;TaskHub=default;Authentication=None",
"AZURE_OPENAI_ENDPOINT": "<AZURE_OPENAI_ENDPOINT>",
"AZURE_OPENAI_DEPLOYMENT_NAME": "<AZURE_OPENAI_DEPLOYMENT_NAME>"
}
}
@@ -1,10 +0,0 @@
{
"IsEncrypted": false,
"Values": {
"FUNCTIONS_WORKER_RUNTIME": "dotnet-isolated",
"AzureWebJobsStorage": "UseDevelopmentStorage=true",
"DURABLE_TASK_SCHEDULER_CONNECTION_STRING": "Endpoint=http://localhost:8080;TaskHub=default;Authentication=None",
"AZURE_OPENAI_ENDPOINT": "<AZURE_OPENAI_ENDPOINT>",
"AZURE_OPENAI_DEPLOYMENT_NAME": "<AZURE_OPENAI_DEPLOYMENT_NAME>"
}
}
@@ -1,12 +0,0 @@
{
"IsEncrypted": false,
"Values": {
"FUNCTIONS_WORKER_RUNTIME": "dotnet-isolated",
"AzureWebJobsStorage": "UseDevelopmentStorage=true",
"DURABLE_TASK_SCHEDULER_CONNECTION_STRING": "Endpoint=http://localhost:8080;TaskHub=default;Authentication=None",
"AZURE_OPENAI_ENDPOINT": "<AZURE_OPENAI_ENDPOINT>",
"AZURE_OPENAI_DEPLOYMENT_NAME": "<AZURE_OPENAI_DEPLOYMENT_NAME>",
"REDIS_CONNECTION_STRING": "localhost:6379",
"REDIS_STREAM_TTL_MINUTES": "10"
}
}
@@ -1,10 +0,0 @@
{
"IsEncrypted": false,
"Values": {
"FUNCTIONS_WORKER_RUNTIME": "dotnet-isolated",
"AzureWebJobsStorage": "UseDevelopmentStorage=true",
"DURABLE_TASK_SCHEDULER_CONNECTION_STRING": "Endpoint=http://localhost:8080;TaskHub=default;Authentication=None",
"AZURE_OPENAI_ENDPOINT": "<AZURE_OPENAI_ENDPOINT>",
"AZURE_OPENAI_DEPLOYMENT_NAME": "<AZURE_OPENAI_DEPLOYMENT_NAME>"
}
}
@@ -1,10 +0,0 @@
{
"IsEncrypted": false,
"Values": {
"FUNCTIONS_WORKER_RUNTIME": "dotnet-isolated",
"AzureWebJobsStorage": "UseDevelopmentStorage=true",
"DURABLE_TASK_SCHEDULER_CONNECTION_STRING": "Endpoint=http://localhost:8080;TaskHub=default;Authentication=None",
"AZURE_OPENAI_ENDPOINT": "<AZURE_OPENAI_ENDPOINT>",
"AZURE_OPENAI_DEPLOYMENT_NAME": "<AZURE_OPENAI_DEPLOYMENT_NAME>"
}
}
@@ -1,10 +0,0 @@
{
"IsEncrypted": false,
"Values": {
"FUNCTIONS_WORKER_RUNTIME": "dotnet-isolated",
"AzureWebJobsStorage": "UseDevelopmentStorage=true",
"DURABLE_TASK_SCHEDULER_CONNECTION_STRING": "Endpoint=http://localhost:8080;TaskHub=default;Authentication=None",
"AZURE_OPENAI_ENDPOINT": "<AZURE_OPENAI_ENDPOINT>",
"AZURE_OPENAI_DEPLOYMENT_NAME": "<AZURE_OPENAI_DEPLOYMENT_NAME>"
}
}
@@ -1,8 +0,0 @@
{
"IsEncrypted": false,
"Values": {
"FUNCTIONS_WORKER_RUNTIME": "dotnet-isolated",
"AzureWebJobsStorage": "UseDevelopmentStorage=true",
"DURABLE_TASK_SCHEDULER_CONNECTION_STRING": "Endpoint=http://localhost:8080;TaskHub=default;Authentication=None"
}
}
@@ -1,10 +0,0 @@
{
"IsEncrypted": false,
"Values": {
"FUNCTIONS_WORKER_RUNTIME": "dotnet-isolated",
"AzureWebJobsStorage": "UseDevelopmentStorage=true",
"DURABLE_TASK_SCHEDULER_CONNECTION_STRING": "Endpoint=http://localhost:8080;TaskHub=default;Authentication=None",
"AZURE_OPENAI_ENDPOINT": "<AZURE_OPENAI_ENDPOINT>",
"AZURE_OPENAI_DEPLOYMENT_NAME": "<AZURE_OPENAI_DEPLOYMENT_NAME>"
}
}
@@ -13,7 +13,7 @@
<ItemGroup>
<PackageReference Include="Azure.AI.Projects" />
<PackageReference Include="Azure.Identity" />
<PackageReference Include="ModelContextProtocol" VersionOverride="1.2.0" />
<PackageReference Include="ModelContextProtocol" />
<PackageReference Include="DotNetEnv" />
</ItemGroup>
@@ -0,0 +1,6 @@
AZURE_AI_PROJECT_ENDPOINT=<your-azure-ai-project-endpoint>
ASPNETCORE_URLS=http://+:8088
ASPNETCORE_ENVIRONMENT=Development
AZURE_AI_MODEL_DEPLOYMENT_NAME=gpt-5
FOUNDRY_TOOLBOX_NAME=<your-toolbox-name>
AZURE_BEARER_TOKEN=DefaultAzureCredential
@@ -0,0 +1,26 @@
# Dockerfile for end-users consuming the Agent Framework via NuGet packages.
#
# This Dockerfile performs a full `dotnet restore` and `dotnet publish` inside the container,
# which only succeeds when the project references its dependencies via PackageReference (see the
# commented-out section in HostedToolboxMcpSkills.csproj). Contributors building from the
# agent-framework repository source must use Dockerfile.contributor instead because
# ProjectReference dependencies live outside this folder and cannot be restored from inside
# this build context.
#
# Use the official .NET 10.0 ASP.NET runtime as a parent image
FROM mcr.microsoft.com/dotnet/aspnet:10.0 AS base
WORKDIR /app
FROM mcr.microsoft.com/dotnet/sdk:10.0 AS build
WORKDIR /src
COPY . .
RUN dotnet restore
RUN dotnet publish -c Release -o /app/publish
# Final stage
FROM base AS final
WORKDIR /app
COPY --from=build /app/publish .
EXPOSE 8088
ENV ASPNETCORE_URLS=http://+:8088
ENTRYPOINT ["dotnet", "HostedToolboxMcpSkills.dll"]
@@ -0,0 +1,18 @@
# Dockerfile for contributors building from the agent-framework repository source.
#
# This project uses ProjectReference to the local source, which means a standard
# multi-stage Docker build cannot resolve dependencies outside this folder.
# Pre-publish the app targeting the container runtime and copy the output:
#
# dotnet publish -c Debug -f net10.0 -r linux-musl-x64 --self-contained false -o out
# docker build -f Dockerfile.contributor -t hosted-toolbox-mcp-skills .
# docker run --rm -p 8088:8088 -e AGENT_NAME=hosted-toolbox-mcp-skills -e AZURE_BEARER_TOKEN=$AZURE_BEARER_TOKEN --env-file .env hosted-toolbox-mcp-skills
#
# For end-users consuming the NuGet package (not ProjectReference), use the standard
# Dockerfile which performs a full dotnet restore + publish inside the container.
FROM mcr.microsoft.com/dotnet/aspnet:10.0-alpine AS final
WORKDIR /app
COPY out/ .
EXPOSE 8088
ENV ASPNETCORE_URLS=http://+:8088
ENTRYPOINT ["dotnet", "HostedToolboxMcpSkills.dll"]
@@ -0,0 +1,36 @@
<Project Sdk="Microsoft.NET.Sdk.Web">
<PropertyGroup>
<TargetFrameworks>net10.0</TargetFrameworks>
<Nullable>enable</Nullable>
<ImplicitUsings>enable</ImplicitUsings>
<CentralPackageTransitivePinningEnabled>false</CentralPackageTransitivePinningEnabled>
<RootNamespace>HostedToolboxMcpSkills</RootNamespace>
<AssemblyName>HostedToolboxMcpSkills</AssemblyName>
<NoWarn>$(NoWarn);</NoWarn>
</PropertyGroup>
<ItemGroup>
<PackageReference Include="Azure.AI.Projects" />
<PackageReference Include="Azure.Identity" />
<PackageReference Include="ModelContextProtocol" VersionOverride="1.2.0" />
<PackageReference Include="DotNetEnv" />
</ItemGroup>
<!-- For contributors: uses ProjectReference to build against local source -->
<ItemGroup>
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.Foundry\Microsoft.Agents.AI.Foundry.csproj" />
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.Foundry.Hosting\Microsoft.Agents.AI.Foundry.Hosting.csproj" />
<ProjectReference Include="..\..\..\..\..\src\Microsoft.Agents.AI.Mcp\Microsoft.Agents.AI.Mcp.csproj" />
<ProjectReference Include="..\Hosted_Shared_Contributor_Setup\Hosted_Shared_Contributor_Setup.csproj" />
</ItemGroup>
<!-- For end-users: uncomment the PackageReference below and remove the ProjectReference above
<ItemGroup>
<PackageReference Include="Microsoft.Agents.AI.Foundry" Version="1.6.1-preview.260514.1" />
<PackageReference Include="Microsoft.Agents.AI.Foundry.Hosting" Version="1.6.1-preview.260514.1" />
<PackageReference Include="Microsoft.Agents.AI.Mcp" Version="1.6.1-preview.260514.1" />
</ItemGroup>
-->
</Project>
@@ -0,0 +1,109 @@
// Copyright (c) Microsoft. All rights reserved.
// Hosted Toolbox MCP Skills Agent
//
// Demonstrates how to host an agent that discovers MCP-based skills from a
// Foundry Toolbox MCP endpoint and injects them as AIContextProviders using
// AgentSkillsProviderBuilder.UseMcpSkills().
//
// Required environment variables:
// AZURE_AI_PROJECT_ENDPOINT - Azure AI Foundry project endpoint
// FOUNDRY_TOOLBOX_NAME - Name of the Foundry Toolbox to connect to
// AZURE_AI_MODEL_DEPLOYMENT_NAME - Model deployment name (default: gpt-5)
using System.Net.Http.Headers;
using Azure.AI.Projects;
using Azure.Core;
using Azure.Identity;
using DotNetEnv;
using Hosted_Shared_Contributor_Setup;
using Microsoft.Agents.AI;
using Microsoft.Agents.AI.Foundry.Hosting;
using ModelContextProtocol.Client;
// Load .env file if present (for local development)
Env.TraversePath().Load();
var projectEndpoint = Environment.GetEnvironmentVariable("AZURE_AI_PROJECT_ENDPOINT")
?? throw new InvalidOperationException("AZURE_AI_PROJECT_ENDPOINT is not set.");
var deployment = Environment.GetEnvironmentVariable("AZURE_AI_MODEL_DEPLOYMENT_NAME") ?? "gpt-5";
var toolboxName = Environment.GetEnvironmentVariable("FOUNDRY_TOOLBOX_NAME")
?? throw new InvalidOperationException("FOUNDRY_TOOLBOX_NAME is not set.");
// Build the Toolbox MCP URL from the project endpoint and toolbox name.
var toolboxMcpServerUrl = $"{projectEndpoint.TrimEnd('/')}/toolboxes/{toolboxName}/mcp?api-version=v1";
// Use a chained credential: try a temporary dev token first (for local Docker debugging),
// then fall back to DefaultAzureCredential (for local dev via dotnet run / managed identity in production).
TokenCredential credential = new ChainedTokenCredential(
new DevTemporaryTokenCredential(),
new DefaultAzureCredential());
// ── Connect to the Foundry Toolbox MCP endpoint ─────────────────────────────
// Create an HttpClient that attaches a fresh Foundry bearer token to every request.
using var httpClient = new HttpClient(new BearerTokenHandler(credential, "https://ai.azure.com/.default") { CheckCertificateRevocationList = true });
Console.WriteLine($"Connecting to Foundry Toolbox '{toolboxName}' MCP server...");
await using var mcpClient = await McpClient.CreateAsync(
new HttpClientTransport(
new HttpClientTransportOptions
{
Endpoint = new Uri(toolboxMcpServerUrl),
Name = toolboxName,
TransportMode = HttpTransportMode.StreamableHttp,
AdditionalHeaders = new Dictionary<string, string>
{
["Foundry-Features"] = "Toolboxes=V1Preview",
},
},
httpClient));
// ── Configure MCP-based skills provider ──────────────────────────────────────
var skillsProvider = new AgentSkillsProviderBuilder()
.UseMcpSkills(mcpClient)
.Build();
// ── Create the agent ─────────────────────────────────────────────────────────
AIAgent agent = new AIProjectClient(new Uri(projectEndpoint), credential)
.AsAIAgent(new ChatClientAgentOptions
{
Name = Environment.GetEnvironmentVariable("AGENT_NAME") ?? "hosted-toolbox-mcp-skills",
Description = "Hosted agent with MCP skills discovered from a Foundry Toolbox",
ChatOptions = new()
{
ModelId = deployment,
Instructions = "You are a helpful assistant.",
},
AIContextProviders = [skillsProvider],
});
// ── Build the host ───────────────────────────────────────────────────────────
var builder = WebApplication.CreateBuilder(args);
builder.Services.AddFoundryResponses(agent);
builder.Services.AddDevTemporaryLocalContributorSetup(); // Local Docker debugging only - must not be used in production.
var app = builder.Build();
app.MapFoundryResponses();
// Contributor-only: in Development, also map the per-agent OpenAI route shape that live Foundry uses
// so a local REPL client can target this server via AIProjectClient.AsAIAgent(Uri agentEndpoint).
// Do not use this in production. Hosted Foundry agents only support the agent-endpoint path.
app.MapDevTemporaryLocalAgentEndpoint();
app.Run();
// ---------------------------------------------------------------------------
// HttpClientHandler: attaches a fresh Foundry bearer token to every request
// ---------------------------------------------------------------------------
internal sealed class BearerTokenHandler(TokenCredential credential, string scope) : HttpClientHandler
{
private readonly TokenRequestContext _tokenContext = new([scope]);
protected override async Task<HttpResponseMessage> SendAsync(HttpRequestMessage request, CancellationToken cancellationToken)
{
AccessToken token = await credential.GetTokenAsync(this._tokenContext, cancellationToken).ConfigureAwait(false);
request.Headers.Authorization = new AuthenticationHeaderValue("Bearer", token.Token);
return await base.SendAsync(request, cancellationToken).ConfigureAwait(false);
}
}
@@ -0,0 +1,103 @@
# Hosted-ToolboxMcpSkills
A hosted agent that discovers **MCP-based skills from a Foundry Toolbox** and makes them available to the agent using `AgentSkillsProviderBuilder.UseMcpSkills(mcpClient)`.
The `AgentSkillsProvider` is attached to the agent as a context provider and implements the [Agent Skills](https://agentskills.io/) progressive-disclosure pattern. When the agent is prompted, it discovers available skills in the Foundry Toolbox via the provider:
1. **Advertise** - skill names and descriptions are injected into the system prompt so the agent knows what is available.
2. **Load** - when the agent decides a skill is relevant, it retrieves the full skill body with detailed instructions via the provider.
3. **Read resources** - if a skill includes supplementary content (reference documents, assets), the agent reads them on demand via the provider.
This way the full skill body and resources are only loaded when the agent actually needs them, reducing token usage.
## Prerequisites
- [.NET 10 SDK](https://dotnet.microsoft.com/download/dotnet/10.0)
- An Azure AI Foundry project with a deployed model (e.g., `gpt-5`)
- A Foundry Toolbox already configured with skills provisioned
- Azure CLI logged in (`az login`)
## Configuration
Copy the template and fill in your values:
```bash
cp .env.example .env
```
Edit `.env` and set your Azure AI Foundry project endpoint and toolbox name:
```env
AZURE_AI_PROJECT_ENDPOINT=https://<your-account>.services.ai.azure.com/api/projects/<your-project>
ASPNETCORE_URLS=http://+:8088
ASPNETCORE_ENVIRONMENT=Development
AZURE_AI_MODEL_DEPLOYMENT_NAME=gpt-5
FOUNDRY_TOOLBOX_NAME=my-toolbox
```
> **Note:** `.env` is gitignored. The `.env.example` template is checked in as a reference.
## Running directly (contributors)
This project uses `ProjectReference` to build against the local Agent Framework source.
```bash
cd dotnet/samples/04-hosting/FoundryHostedAgents/responses/Hosted-ToolboxMcpSkills
dotnet run
```
The agent will start on `http://localhost:8088`.
### Test it
Using the Azure Developer CLI:
```bash
azd ai agent invoke --local "What skills do you have available?"
```
## Running with Docker
Since this project uses `ProjectReference`, use `Dockerfile.contributor` which takes a pre-published output.
### 1. Publish for the container runtime (Linux Alpine)
```bash
dotnet publish -c Debug -f net10.0 -r linux-musl-x64 --self-contained false -o out
```
### 2. Build the Docker image
```bash
docker build -f Dockerfile.contributor -t hosted-toolbox-mcp-skills .
```
### 3. Run the container
Generate a bearer token on your host and pass it to the container:
```bash
# Generate token (expires in ~1 hour)
export AZURE_BEARER_TOKEN=$(az account get-access-token --resource https://ai.azure.com --query accessToken -o tsv)
# Run with token
docker run --rm -p 8088:8088 \
-e AGENT_NAME=hosted-toolbox-mcp-skills \
-e AZURE_BEARER_TOKEN=$AZURE_BEARER_TOKEN \
--env-file .env \
hosted-toolbox-mcp-skills
```
> **Note:** `AGENT_NAME` is passed via `-e` to simulate the platform injection. `AZURE_BEARER_TOKEN` provides Azure credentials to the container (tokens expire after ~1 hour). The `.env` file provides the remaining configuration.
### 4. Test it
Using the Azure Developer CLI:
```bash
azd ai agent invoke --local "What skills do you have available?"
```
## NuGet package users
If you are consuming the Agent Framework as a NuGet package (not building from source), use the standard `Dockerfile` instead of `Dockerfile.contributor`. See the commented section in `HostedToolboxMcpSkills.csproj` for the `PackageReference` alternative.
@@ -0,0 +1,43 @@
# yaml-language-server: $schema=https://raw.githubusercontent.com/microsoft/AgentSchema/refs/heads/main/schemas/v1.0/AgentManifest.yaml
name: hosted-toolbox-mcp-skills
displayName: "Hosted Toolbox MCP Skills Agent"
description: >
A hosted agent that discovers MCP-based skills from a Foundry Toolbox
and makes them available to the agent via the agent skills provider.
metadata:
tags:
- AI Agent Hosting
- Azure AI AgentServer
- Responses Protocol
- Agent Framework
- MCP
- Model Context Protocol
- Agent Skills
- Foundry Toolbox
- Foundry Toolbox Skills
template:
name: hosted-toolbox-mcp-skills
kind: hosted
protocols:
- protocol: responses
version: 1.0.0
resources:
cpu: "0.25"
memory: 0.5Gi
environment_variables:
- name: AZURE_AI_MODEL_DEPLOYMENT_NAME
value: "{{AZURE_AI_MODEL_DEPLOYMENT_NAME}}"
- name: FOUNDRY_TOOLBOX_NAME
value: "{{FOUNDRY_TOOLBOX_NAME}}"
parameters:
properties:
- name: FOUNDRY_TOOLBOX_NAME
secret: false
description: Name of the Foundry Toolbox to connect to for MCP skill discovery
resources:
- kind: model
id: gpt-5
name: AZURE_AI_MODEL_DEPLOYMENT_NAME
@@ -0,0 +1,14 @@
# yaml-language-server: $schema=https://raw.githubusercontent.com/microsoft/AgentSchema/refs/heads/main/schemas/v1.0/ContainerAgent.yaml
kind: hosted
name: hosted-toolbox-mcp-skills
protocols:
- protocol: responses
version: 1.0.0
resources:
cpu: "0.25"
memory: 0.5Gi
environment_variables:
- name: AZURE_AI_MODEL_DEPLOYMENT_NAME
value: ${AZURE_AI_MODEL_DEPLOYMENT_NAME}
- name: FOUNDRY_TOOLBOX_NAME
value: ${FOUNDRY_TOOLBOX_NAME}
@@ -15,6 +15,7 @@
<ItemGroup>
<ProjectReference Include="..\..\..\..\src\Microsoft.Agents.AI.Hosting.AGUI.AspNetCore\Microsoft.Agents.AI.Hosting.AGUI.AspNetCore.csproj" />
<ProjectReference Include="..\..\..\..\src\Microsoft.Agents.AI.Hosting.AspNetCore\Microsoft.Agents.AI.Hosting.AspNetCore.csproj" />
<ProjectReference Include="..\..\..\..\src\Microsoft.Agents.AI.OpenAI\Microsoft.Agents.AI.OpenAI.csproj" />
</ItemGroup>
@@ -19,6 +19,11 @@ builder.Services.AddHttpClient().AddLogging();
builder.Services.ConfigureHttpJsonOptions(options => options.SerializerOptions.TypeInfoResolverChain.Add(AGUIDojoServerSerializerContext.Default));
builder.Services.AddAGUI();
// WARNING: When adding session persistence (e.g., WithInMemorySessionStore), or running in production,
// make sure to also register a SessionIsolationKeyProvider to scope sessions by principal in multi-user
// deployments, e.g.:
// builder.Services.UseClaimsBasedSessionIsolation(new() { ClaimType = ClaimTypes.NameIdentifier });
WebApplication app = builder.Build();
app.UseHttpLogging();
@@ -49,8 +49,9 @@ var agent = new AzureOpenAIClient(
AGUIServerSerializerContext.Default.Options)
]);
// When running in production, make sure to use an SessionIsolationKeyProvider, e.g. ClaimsIdentity-based
// if using Claims-based Identity for Authentication/Authorization
// WARNING: When adding session persistence (e.g., WithInMemorySessionStore), or running in production,
// make sure to also register a SessionIsolationKeyProvider to scope sessions by principal in multi-user
// deployments, e.g.:
// builder.Services.UseClaimsBasedSessionIsolation(new() { ClaimType = ClaimTypes.NameIdentifier });
// Register the agent with the host and configure it to use an in-memory session store
@@ -14,6 +14,7 @@
<ItemGroup>
<ProjectReference Include="..\..\..\..\src\Microsoft.Agents.AI.Hosting.AGUI.AspNetCore\Microsoft.Agents.AI.Hosting.AGUI.AspNetCore.csproj" />
<ProjectReference Include="..\..\..\..\src\Microsoft.Agents.AI.Hosting.AspNetCore\Microsoft.Agents.AI.Hosting.AspNetCore.csproj" />
<ProjectReference Include="..\..\..\..\src\Microsoft.Agents.AI.AGUI\Microsoft.Agents.AI.AGUI.csproj" />
<ProjectReference Include="..\..\..\..\src\Microsoft.Agents.AI.OpenAI\Microsoft.Agents.AI.OpenAI.csproj" />
</ItemGroup>
@@ -12,6 +12,11 @@ WebApplicationBuilder builder = WebApplication.CreateBuilder(args);
builder.Services.AddHttpClient().AddLogging();
builder.Services.AddAGUI();
// WARNING: When adding session persistence (e.g., WithInMemorySessionStore), or running in production,
// make sure to also register a SessionIsolationKeyProvider to scope sessions by principal in multi-user
// deployments, e.g.:
// builder.Services.UseClaimsBasedSessionIsolation(new() { ClaimType = ClaimTypes.NameIdentifier });
WebApplication app = builder.Build();
string endpoint = builder.Configuration["AZURE_OPENAI_ENDPOINT"] ?? throw new InvalidOperationException("AZURE_OPENAI_ENDPOINT is not set.");
@@ -24,11 +24,13 @@
<ItemGroup>
<PackageReference Include="Azure.AI.Projects" />
<PackageReference Include="Azure.Core" />
<PackageReference Include="Microsoft.Extensions.AI" />
<PackageReference Include="Microsoft.Extensions.AI.Abstractions" />
<PackageReference Include="Microsoft.Extensions.AI.OpenAI" />
<PackageReference Include="Microsoft.Extensions.Compliance.Abstractions" />
<PackageReference Include="OpenAI" />
<PackageReference Include="System.ClientModel" />
</ItemGroup>
<!-- Evaluation support requires net8.0+ (MEAI.Evaluation does not support legacy TFMs) -->
@@ -1,7 +1,9 @@
// Copyright (c) Microsoft. All rights reserved.
using System;
using System.Diagnostics.CodeAnalysis;
using Microsoft.Agents.AI;
using Microsoft.Extensions.Logging;
using Microsoft.Shared.DiagnosticIds;
namespace Microsoft.Extensions.AI;
@@ -32,11 +34,19 @@ public static class ChatClientHarnessExtensions
/// additional context providers, and chat history provider.
/// When <see langword="null"/>, the agent uses built-in default settings.
/// </param>
/// <param name="loggerFactory">
/// Optional logger factory for creating loggers used by the agent and its components.
/// </param>
/// <param name="services">
/// Optional service provider for resolving dependencies required by AI functions and other agent components.
/// </param>
/// <returns>A new <see cref="HarnessAgent"/> instance.</returns>
public static HarnessAgent AsHarnessAgent(
this IChatClient chatClient,
int maxContextWindowTokens,
int maxOutputTokens,
HarnessAgentOptions? options = null) =>
new(chatClient, maxContextWindowTokens, maxOutputTokens, options);
HarnessAgentOptions? options = null,
ILoggerFactory? loggerFactory = null,
IServiceProvider? services = null) =>
new(chatClient, maxContextWindowTokens, maxOutputTokens, options, loggerFactory, services);
}
@@ -10,6 +10,7 @@ using Microsoft.Agents.AI.Compaction;
using Microsoft.Agents.AI.Tools.Shell;
#endif
using Microsoft.Extensions.AI;
using Microsoft.Extensions.Logging;
using Microsoft.Shared.DiagnosticIds;
using Microsoft.Shared.Diagnostics;
@@ -105,6 +106,12 @@ public sealed class HarnessAgent : DelegatingAIAgent
/// additional context providers, and chat history provider.
/// When <see langword="null"/>, the agent uses built-in default settings.
/// </param>
/// <param name="loggerFactory">
/// Optional logger factory for creating loggers used by the agent and its components.
/// </param>
/// <param name="services">
/// Optional service provider for resolving dependencies required by AI functions and other agent components.
/// </param>
/// <exception cref="ArgumentNullException">
/// <paramref name="chatClient"/> is <see langword="null"/>.
/// </exception>
@@ -112,18 +119,20 @@ public sealed class HarnessAgent : DelegatingAIAgent
/// <paramref name="maxContextWindowTokens"/> is not positive, or
/// <paramref name="maxOutputTokens"/> is negative or greater than or equal to <paramref name="maxContextWindowTokens"/>.
/// </exception>
public HarnessAgent(IChatClient chatClient, int maxContextWindowTokens, int maxOutputTokens, HarnessAgentOptions? options = null)
public HarnessAgent(IChatClient chatClient, int maxContextWindowTokens, int maxOutputTokens, HarnessAgentOptions? options = null, ILoggerFactory? loggerFactory = null, IServiceProvider? services = null)
: base(BuildAgent(
Throw.IfNull(chatClient),
maxContextWindowTokens,
maxOutputTokens,
options))
options,
loggerFactory,
services))
{
}
private static AIAgent BuildAgent(IChatClient chatClient, int maxContextWindowTokens, int maxOutputTokens, HarnessAgentOptions? options)
private static AIAgent BuildAgent(IChatClient chatClient, int maxContextWindowTokens, int maxOutputTokens, HarnessAgentOptions? options, ILoggerFactory? loggerFactory, IServiceProvider? services)
{
ChatClientAgent innerAgent = BuildInnerAgent(chatClient, maxContextWindowTokens, maxOutputTokens, options);
ChatClientAgent innerAgent = BuildInnerAgent(chatClient, maxContextWindowTokens, maxOutputTokens, options, loggerFactory, services);
AIAgentBuilder builder = innerAgent.AsBuilder();
@@ -137,10 +146,10 @@ public sealed class HarnessAgent : DelegatingAIAgent
builder.UseOpenTelemetry(sourceName: options?.OpenTelemetrySourceName);
}
return builder.Build();
return builder.Build(services);
}
private static ChatClientAgent BuildInnerAgent(IChatClient chatClient, int maxContextWindowTokens, int maxOutputTokens, HarnessAgentOptions? options)
private static ChatClientAgent BuildInnerAgent(IChatClient chatClient, int maxContextWindowTokens, int maxOutputTokens, HarnessAgentOptions? options, ILoggerFactory? loggerFactory, IServiceProvider? services)
{
var compactionStrategy = new ContextWindowCompactionStrategy(
maxContextWindowTokens: maxContextWindowTokens,
@@ -165,13 +174,13 @@ public sealed class HarnessAgent : DelegatingAIAgent
ChatOptions chatOptions = BuildChatOptions(options, instructions, maxOutputTokens);
var compactionProvider = new CompactionProvider(compactionStrategy);
var compactionProvider = new CompactionProvider(compactionStrategy, loggerFactory: loggerFactory);
IEnumerable<AIContextProvider> contextProviders = BuildContextProviders(options);
IEnumerable<AIContextProvider> contextProviders = BuildContextProviders(options, loggerFactory);
return chatClient
.AsBuilder()
.UseFunctionInvocation(configure: options?.MaximumIterationsPerRequest is int maxIterations
.UseFunctionInvocation(loggerFactory, configure: options?.MaximumIterationsPerRequest is int maxIterations
? ficc => ficc.MaximumIterationsPerRequest = maxIterations
: null)
.UseMessageInjection()
@@ -189,7 +198,9 @@ public sealed class HarnessAgent : DelegatingAIAgent
RequirePerServiceCallChatHistoryPersistence = true,
WarnOnChatHistoryProviderConflict = false,
ThrowOnChatHistoryProviderConflict = false,
});
},
loggerFactory,
services);
}
private static ChatOptions BuildChatOptions(HarnessAgentOptions? options, string instructions, int maxOutputTokens)
@@ -215,7 +226,7 @@ public sealed class HarnessAgent : DelegatingAIAgent
return result;
}
private static List<AIContextProvider> BuildContextProviders(HarnessAgentOptions? options)
private static List<AIContextProvider> BuildContextProviders(HarnessAgentOptions? options, ILoggerFactory? loggerFactory)
{
var providers = new List<AIContextProvider>();
@@ -255,8 +266,8 @@ public sealed class HarnessAgent : DelegatingAIAgent
if (options?.DisableAgentSkillsProvider is not true)
{
AgentSkillsProvider skillsProvider = options?.AgentSkillsSource is AgentSkillsSource source
? new AgentSkillsProvider(source)
: new AgentSkillsProvider(Directory.GetCurrentDirectory());
? new AgentSkillsProvider(source, loggerFactory: loggerFactory)
: new AgentSkillsProvider(Directory.GetCurrentDirectory(), loggerFactory: loggerFactory);
providers.Add(skillsProvider);
}
@@ -103,7 +103,16 @@ public static class AGUIEndpointRouteBuilderExtensions
ArgumentNullException.ThrowIfNull(aiAgent);
var agentSessionStore = endpoints.ServiceProvider.GetKeyedService<AgentSessionStore>(aiAgent.Name);
var hostAgent = new AIHostAgent(aiAgent, agentSessionStore ?? new NoopAgentSessionStore());
// Ensure that we have an IsolationKeyScopedAgentSessionStore registered.
var isolationKeyProvider = endpoints.ServiceProvider.GetService<SessionIsolationKeyProvider>();
if (agentSessionStore?.GetService<IsolationKeyScopedAgentSessionStore>() is null)
{
agentSessionStore ??= new NoopAgentSessionStore();
agentSessionStore = new IsolationKeyScopedAgentSessionStore(agentSessionStore, isolationKeyProvider, new() { Strict = isolationKeyProvider != null });
}
var hostAgent = new AIHostAgent(aiAgent, agentSessionStore);
return endpoints.MapPost(pattern, async ([FromBody] RunAgentInput? input, HttpContext context, CancellationToken cancellationToken) =>
{
@@ -1,7 +1,7 @@
<Project Sdk="Microsoft.NET.Sdk">
<PropertyGroup>
<IsReleaseCandidate>true</IsReleaseCandidate>
<!-- Preview while Microsoft.Agents.AI.Foundry is preview (blocked by Azure.AI.Projects 2.1.0-beta). Flip to IsReleased=true once that ships stable. -->
<NoWarn>$(NoWarn);MEAI001;OPENAI001</NoWarn>
</PropertyGroup>
@@ -1,7 +1,7 @@
<Project Sdk="Microsoft.NET.Sdk">
<PropertyGroup>
<IsReleaseCandidate>true</IsReleaseCandidate>
<IsReleased>true</IsReleased>
<NoWarn>$(NoWarn);MEAI001;OPENAI001</NoWarn>
</PropertyGroup>
@@ -13,9 +13,11 @@
<Import Project="$(RepoRoot)/dotnet/nuget/nuget-package.props" />
<!-- Package not yet published to NuGet — disable baseline validation until first release -->
<!-- First Stable release after the RC milestone. Baseline against the latest
published RC so package validation catches accidental breaking changes.
Future releases should bump this to the previous stable version. -->
<PropertyGroup>
<EnablePackageValidation>false</EnablePackageValidation>
<PackageValidationBaselineVersion>1.8.0-rc1</PackageValidationBaselineVersion>
</PropertyGroup>
<PropertyGroup>
@@ -1,7 +1,7 @@
<Project Sdk="Microsoft.NET.Sdk">
<PropertyGroup>
<IsReleaseCandidate>true</IsReleaseCandidate>
<IsReleased>true</IsReleased>
<NoWarn>$(NoWarn);MEAI001;OPENAI001</NoWarn>
</PropertyGroup>
@@ -13,6 +13,13 @@
<Import Project="$(RepoRoot)/dotnet/nuget/nuget-package.props" />
<!-- First Stable release after the RC milestone. Baseline against the latest
published RC so package validation catches accidental breaking changes.
Future releases should bump this to the previous stable version. -->
<PropertyGroup>
<PackageValidationBaselineVersion>1.8.0-rc1</PackageValidationBaselineVersion>
</PropertyGroup>
<PropertyGroup>
<!-- NuGet Package Settings -->
<Title>Microsoft Agent Framework Declarative Workflows</Title>
@@ -27,6 +27,14 @@ internal sealed class InvokeMcpToolExecutor(
WorkflowFormulaState state) :
DeclarativeActionExecutor<InvokeMcpTool>(model, state)
{
private const string ApprovalSnapshotStateKey = nameof(_approvalSnapshot);
/// <summary>
/// Snapshot of evaluated parameters at approval-request time.
/// Used to prevent TOCTOU attacks where state mutates during the approval window.
/// </summary>
private ApprovalSnapshot? _approvalSnapshot;
/// <summary>
/// Step identifiers for the MCP tool invocation workflow.
/// </summary>
@@ -75,6 +83,10 @@ internal sealed class InvokeMcpToolExecutor(
if (requireApproval)
{
// Snapshot the evaluated parameters to prevent TOCTOU attacks.
// If state mutates during the approval window, the approved values are used on resume.
this._approvalSnapshot = new ApprovalSnapshot(serverUrl, serverLabel, toolName, arguments, connectionName);
// Create tool call content for approval request.
// Transport headers (e.g. Authorization) are intentionally excluded from the
// approval event: they must not cross into the externally-surfaced approval request.
@@ -137,13 +149,14 @@ internal sealed class InvokeMcpToolExecutor(
return;
}
// Approved - now invoke the tool
string serverUrl = this.GetServerUrl();
string? serverLabel = this.GetServerLabel();
string toolName = this.GetToolName();
Dictionary<string, object?>? arguments = this.GetArguments();
// Approved - use the snapshot from approval-request time to prevent TOCTOU attacks.
// Headers are re-evaluated (they may contain auth secrets that should not be persisted).
string serverUrl = this._approvalSnapshot?.ServerUrl ?? this.GetServerUrl();
string? serverLabel = this._approvalSnapshot?.ServerLabel ?? this.GetServerLabel();
string toolName = this._approvalSnapshot?.ToolName ?? this.GetToolName();
Dictionary<string, object?>? arguments = this._approvalSnapshot?.Arguments ?? this.GetArguments();
Dictionary<string, string>? headers = this.GetHeaders();
string? connectionName = this.GetConnectionName();
string? connectionName = this._approvalSnapshot?.ConnectionName ?? this.GetConnectionName();
McpServerToolResultContent resultContent = await mcpToolHandler.InvokeToolAsync(
serverUrl,
@@ -162,9 +175,33 @@ internal sealed class InvokeMcpToolExecutor(
/// </summary>
public async ValueTask CompleteAsync(IWorkflowContext context, ActionExecutorResult message, CancellationToken cancellationToken)
{
// Clear the approval snapshot after successful completion.
this._approvalSnapshot = null;
await ClearSnapshotStateAsync(context, cancellationToken).ConfigureAwait(false);
await context.RaiseCompletionEventAsync(this.Model, cancellationToken).ConfigureAwait(false);
}
/// <inheritdoc/>
/// <remarks>
/// Persists the approval snapshot to workflow state so it survives checkpoint/restore cycles.
/// </remarks>
protected override async ValueTask OnCheckpointingAsync(IWorkflowContext context, CancellationToken cancellationToken = default)
{
await context.QueueStateUpdateAsync(ApprovalSnapshotStateKey, this._approvalSnapshot, null, cancellationToken).ConfigureAwait(false);
await base.OnCheckpointingAsync(context, cancellationToken).ConfigureAwait(false);
}
/// <inheritdoc/>
/// <remarks>
/// Restores the approval snapshot from workflow state after a checkpoint restore.
/// </remarks>
protected override async ValueTask OnCheckpointRestoredAsync(IWorkflowContext context, CancellationToken cancellationToken = default)
{
await base.OnCheckpointRestoredAsync(context, cancellationToken).ConfigureAwait(false);
this._approvalSnapshot = await context.ReadStateAsync<ApprovalSnapshot>(ApprovalSnapshotStateKey, null, cancellationToken).ConfigureAwait(false);
}
private async ValueTask ProcessResultAsync(IWorkflowContext context, McpServerToolResultContent resultContent, CancellationToken cancellationToken)
{
bool autoSend = this.GetAutoSendValue();
@@ -365,4 +402,24 @@ internal sealed class InvokeMcpToolExecutor(
return result;
}
/// <summary>
/// Clears the persisted approval snapshot state after a successful tool invocation.
/// </summary>
private static async ValueTask ClearSnapshotStateAsync(IWorkflowContext context, CancellationToken cancellationToken)
{
await context.QueueStateUpdateAsync<ApprovalSnapshot?>(ApprovalSnapshotStateKey, null, null, cancellationToken).ConfigureAwait(false);
}
/// <summary>
/// Stores the evaluated parameters at approval-request time so that
/// <see cref="CaptureResponseAsync"/> uses the values the user reviewed,
/// even if <see cref="WorkflowFormulaState"/> mutates during the approval window.
/// </summary>
internal sealed record ApprovalSnapshot(
string ServerUrl,
string? ServerLabel,
string ToolName,
Dictionary<string, object?>? Arguments,
string? ConnectionName);
}
@@ -49,14 +49,13 @@ public sealed class AgentFileSkill : AgentSkill
/// <inheritdoc/>
/// <remarks>
/// Returns the raw SKILL.md content. When the skill has scripts, a
/// <c>&lt;scripts&gt;&lt;script name="..."&gt;&lt;parameters_schema&gt;...&lt;/parameters_schema&gt;&lt;/script&gt;&lt;/scripts&gt;</c>
/// block is appended with a per-script entry describing the expected argument format.
/// <c>&lt;script_schemas&gt;</c> block is appended describing the argument format.
/// The result is cached after the first access.
/// </remarks>
public override ValueTask<string> GetContentAsync(CancellationToken cancellationToken = default)
{
var content = this._content ??= this._scripts is { Count: > 0 }
? this._originalContent + AgentInlineSkillContentBuilder.BuildScriptsBlock(this._scripts)
? this._originalContent + AgentInlineSkillContentBuilder.BuildScriptSchemasBlock(this._scripts)
: this._originalContent;
return new(content);
}
@@ -114,7 +114,6 @@ public abstract class AgentClassSkill<
this.Frontmatter.Name,
this.Frontmatter.Description,
this.Instructions,
this.Resources,
this.Scripts));
}
@@ -147,11 +146,17 @@ public abstract class AgentClassSkill<
/// Gets the resources associated with this skill, or <see langword="null"/> if none.
/// </summary>
/// <remarks>
/// <para>
/// The default implementation returns resources discovered via reflection by scanning
/// <typeparamref name="TSelf"/> for members annotated with <see cref="AgentSkillResourceAttribute"/>.
/// This discovery is compatible with Native AOT because <typeparamref name="TSelf"/> is annotated with
/// <see cref="DynamicallyAccessedMembersAttribute"/>. The result is cached after the first access.
/// Override this property in derived classes to provide skill-specific resources.
/// </para>
/// <para>
/// Resources are not automatically included in the skill body.
/// To enable discovery, reference resources by name in the skill's instructions or in other resources.
/// </para>
/// </remarks>
public virtual IReadOnlyList<AgentSkillResource>? Resources => this._resources.Value;
@@ -159,11 +164,17 @@ public abstract class AgentClassSkill<
/// Gets the scripts associated with this skill, or <see langword="null"/> if none.
/// </summary>
/// <remarks>
/// <para>
/// The default implementation returns scripts discovered via reflection by scanning
/// <typeparamref name="TSelf"/> for methods annotated with <see cref="AgentSkillScriptAttribute"/>.
/// This discovery is compatible with Native AOT because <typeparamref name="TSelf"/> is annotated with
/// <see cref="DynamicallyAccessedMembersAttribute"/>. The result is cached after the first access.
/// Override this property in derived classes to provide skill-specific scripts.
/// </para>
/// <para>
/// Only script parameter schemas are included in the skill body (as a <c>&lt;script_schemas&gt;</c> block).
/// To enable discovery, reference scripts by name in the skill's instructions or in a resource.
/// </para>
/// </remarks>
public virtual IReadOnlyList<AgentSkillScript>? Scripts => this._scripts.Value;
@@ -184,6 +195,10 @@ public abstract class AgentClassSkill<
/// <summary>
/// Creates a skill resource backed by a static value.
/// </summary>
/// <remarks>
/// Resources are not automatically included in the skill body.
/// To enable discovery, reference the resource by name in the skill's instructions or in another resource.
/// </remarks>
/// <param name="name">The resource name.</param>
/// <param name="value">The static resource value.</param>
/// <param name="description">An optional description of the resource.</param>
@@ -194,6 +209,10 @@ public abstract class AgentClassSkill<
/// <summary>
/// Creates a skill resource backed by a delegate that produces a dynamic value.
/// </summary>
/// <remarks>
/// Resources are not automatically included in the skill body.
/// To enable discovery, reference the resource by name in the skill's instructions or in another resource.
/// </remarks>
/// <param name="name">The resource name.</param>
/// <param name="method">A method that produces the resource value when requested.</param>
/// <param name="description">An optional description of the resource.</param>
@@ -208,6 +227,10 @@ public abstract class AgentClassSkill<
/// <summary>
/// Creates a skill script backed by a delegate.
/// </summary>
/// <remarks>
/// Only the script's parameter schema is included in the skill body (as a <c>&lt;script_schemas&gt;</c> block).
/// To enable discovery, reference the script by name in the skill's instructions or in a resource.
/// </remarks>
/// <param name="name">The script name.</param>
/// <param name="method">A method to execute when the script is invoked.</param>
/// <param name="description">An optional description of the script.</param>
@@ -95,7 +95,7 @@ public sealed class AgentInlineSkill : AgentSkill
/// <inheritdoc/>
public override ValueTask<string> GetContentAsync(CancellationToken cancellationToken = default)
{
return new(this._cachedContent ??= AgentInlineSkillContentBuilder.Build(this.Frontmatter.Name, this.Frontmatter.Description, this._instructions, this._resources, this._scripts));
return new(this._cachedContent ??= AgentInlineSkillContentBuilder.Build(this.Frontmatter.Name, this.Frontmatter.Description, this._instructions, this._scripts));
}
/// <inheritdoc/>
@@ -115,6 +115,10 @@ public sealed class AgentInlineSkill : AgentSkill
/// <summary>
/// Registers a static resource with this skill.
/// </summary>
/// <remarks>
/// Resources are not automatically included in the skill body.
/// To enable discovery, reference the resource by name in the skill's instructions or in another resource.
/// </remarks>
/// <param name="name">The resource name.</param>
/// <param name="value">The static resource value.</param>
/// <param name="description">An optional description of the resource.</param>
@@ -129,6 +133,10 @@ public sealed class AgentInlineSkill : AgentSkill
/// Registers a dynamic resource with this skill, backed by a C# delegate.
/// The delegate's parameters and return type are automatically marshaled via <c>AIFunctionFactory</c>.
/// </summary>
/// <remarks>
/// Resources are not automatically included in the skill body.
/// To enable discovery, reference the resource by name in the skill's instructions or in another resource.
/// </remarks>
/// <param name="name">The resource name.</param>
/// <param name="method">A method that produces the resource value when requested.</param>
/// <param name="description">An optional description of the resource.</param>
@@ -147,6 +155,10 @@ public sealed class AgentInlineSkill : AgentSkill
/// Registers a script with this skill, backed by a C# delegate.
/// The delegate's parameters and return type are automatically marshaled via <c>AIFunctionFactory</c>.
/// </summary>
/// <remarks>
/// Only the script's parameter schema is included in the skill body (as a <c>&lt;script_schemas&gt;</c> block).
/// To enable discovery, reference the script by name in the skill's instructions or in a resource.
/// </remarks>
/// <param name="name">The script name.</param>
/// <param name="method">A method to execute when the script is invoked.</param>
/// <param name="description">An optional description of the script.</param>
@@ -12,19 +12,17 @@ namespace Microsoft.Agents.AI;
internal static class AgentInlineSkillContentBuilder
{
/// <summary>
/// Builds the complete skill content containing name, description, instructions, resources, and scripts.
/// Builds the complete skill content containing name, description, instructions, and script parameter schemas.
/// </summary>
/// <param name="name">The skill name.</param>
/// <param name="description">The skill description.</param>
/// <param name="instructions">The raw instructions text.</param>
/// <param name="resources">Optional resources associated with the skill.</param>
/// <param name="scripts">Optional scripts associated with the skill.</param>
/// <returns>An XML-structured content string.</returns>
public static string Build(
string name,
string description,
string instructions,
IReadOnlyList<AgentSkillResource>? resources,
IReadOnlyList<AgentSkillScript>? scripts)
{
_ = Throw.IfNullOrWhitespace(name);
@@ -39,41 +37,24 @@ internal static class AgentInlineSkillContentBuilder
.Append(EscapeXmlString(instructions))
.Append("\n</instructions>");
if (resources is { Count: > 0 })
{
sb.Append("\n\n<resources>\n");
foreach (var resource in resources)
{
if (resource.Description is not null)
{
sb.Append($" <resource name=\"{EscapeXmlString(resource.Name)}\" description=\"{EscapeXmlString(resource.Description)}\"/>\n");
}
else
{
sb.Append($" <resource name=\"{EscapeXmlString(resource.Name)}\"/>\n");
}
}
sb.Append("</resources>");
}
if (scripts is { Count: > 0 })
{
sb.Append('\n');
sb.Append(BuildScriptsBlock(scripts));
sb.Append(BuildScriptSchemasBlock(scripts));
}
return sb.ToString();
}
/// <summary>
/// Builds a <c>&lt;scripts&gt;...&lt;/scripts&gt;</c> XML block for the given scripts.
/// Each script is emitted as a <c>&lt;script name="..."&gt;</c> element with optional
/// <c>description</c> attribute and <c>&lt;parameters_schema&gt;</c> child element.
/// Builds a <c>&lt;script_schemas&gt;...&lt;/script_schemas&gt;</c> XML block for the given scripts.
/// Each script is emitted as a <c>&lt;schema script="..."&gt;</c> element containing only
/// the parameter schema. This block serves as a reference for the model to know how to
/// format arguments when calling scripts, not as a discovery mechanism.
/// </summary>
/// <param name="scripts">The scripts to include in the block.</param>
/// <returns>An XML string starting with <c>\n&lt;scripts&gt;</c>, or an empty string if the list is empty.</returns>
public static string BuildScriptsBlock(IReadOnlyList<AgentSkillScript> scripts)
/// <returns>An XML string starting with <c>\n&lt;script_schemas&gt;</c>, or an empty string if the list is empty.</returns>
public static string BuildScriptSchemasBlock(IReadOnlyList<AgentSkillScript> scripts)
{
_ = Throw.IfNull(scripts);
@@ -83,32 +64,23 @@ internal static class AgentInlineSkillContentBuilder
}
var sb = new StringBuilder();
sb.Append("\n<scripts>\n");
sb.Append("\n<script_schemas>\n");
foreach (var script in scripts)
{
var parametersSchema = script.ParametersSchema;
if (script.Description is null && parametersSchema is null)
if (parametersSchema is null)
{
sb.Append($" <script name=\"{EscapeXmlString(script.Name)}\"/>\n");
sb.Append($" <schema script=\"{EscapeXmlString(script.Name)}\"/>\n");
}
else
{
sb.Append(script.Description is not null
? $" <script name=\"{EscapeXmlString(script.Name)}\" description=\"{EscapeXmlString(script.Description)}\">\n"
: $" <script name=\"{EscapeXmlString(script.Name)}\">\n");
if (parametersSchema is not null)
{
sb.Append($" <parameters_schema>{EscapeXmlString(parametersSchema.Value.GetRawText(), preserveQuotes: true)}</parameters_schema>\n");
}
sb.Append(" </script>\n");
sb.Append($" <schema script=\"{EscapeXmlString(script.Name)}\">{EscapeXmlString(parametersSchema.Value.GetRawText(), preserveQuotes: true)}</schema>\n");
}
}
sb.Append("</scripts>");
sb.Append("</script_schemas>");
return sb.ToString();
}
@@ -6,11 +6,11 @@
.DESCRIPTION
The IT fixture targets stable, scenario-keyed agent names (e.g. it-happy-path) and only
manages versions on each test run. The agent itself must already exist AND its managed
identity must hold the Azure AI User role on the project scope, otherwise inbound
identity must hold the Foundry User role on the project scope, otherwise inbound
inference calls fail with HTTP 500 PermissionDenied.
This script idempotently creates each scenario agent (with a placeholder version) and
grants Azure AI User on the project to its managed identity. Re-run it safely; existing
grants Foundry User on the project to its managed identity. Re-run it safely; existing
agents and role assignments are left in place.
.PARAMETER ProjectEndpoint
@@ -135,20 +135,20 @@ foreach ($scenario in $Scenarios) {
-Body $patchBody | Out-Null
}
# 3. Grant Azure AI User on the project scope to the agent MI (idempotent).
# 3. Grant Foundry User on the project scope to the agent MI (idempotent).
$existing = az role assignment list --assignee $principalId --scope $projectScope `
--query "[?roleDefinitionName=='Azure AI User']" 2>$null | ConvertFrom-Json
--query "[?roleDefinitionName=='Foundry User']" 2>$null | ConvertFrom-Json
if ($existing) {
Write-Host " role already assigned"
} else {
Write-Host " granting Azure AI User..."
Write-Host " granting Foundry User..."
$maxAttempts = 12
$granted = $false
for ($i = 1; $i -le $maxAttempts; $i++) {
$output = az role assignment create `
--assignee-object-id $principalId `
--assignee-principal-type ServicePrincipal `
--role 'Azure AI User' `
--role 'Foundry User' `
--scope $projectScope 2>&1
if ($LASTEXITCODE -eq 0) {
$granted = $true
@@ -1,12 +0,0 @@
{
"profiles": {
"Microsoft.Agents.AI.DevUI.UnitTests": {
"commandName": "Project",
"launchBrowser": true,
"environmentVariables": {
"ASPNETCORE_ENVIRONMENT": "Development"
},
"applicationUrl": "https://localhost:63009;http://localhost:63010"
}
}
}
@@ -9,6 +9,7 @@ using System.Threading.Tasks;
using Microsoft.Agents.AI.Tools.Shell;
#endif
using Microsoft.Extensions.AI;
using Microsoft.Extensions.Logging;
using Moq;
namespace Microsoft.Agents.AI.UnitTests;
@@ -1460,4 +1461,131 @@ public class HarnessAgentTests
#endregion
#endif
#region LoggerFactory and ServiceProvider
/// <summary>
/// Verify that the constructor succeeds when loggerFactory is provided.
/// </summary>
[Fact]
public void Constructor_SucceedsWithLoggerFactory()
{
// Arrange
var chatClient = new Mock<IChatClient>().Object;
var loggerFactory = new Mock<ILoggerFactory>().Object;
// Act
var agent = new HarnessAgent(chatClient, TestMaxContextWindowTokens, TestMaxOutputTokens, CreateAllDisabledOptions(), loggerFactory);
// Assert
Assert.NotNull(agent);
}
/// <summary>
/// Verify that the constructor succeeds when serviceProvider is provided.
/// </summary>
[Fact]
public void Constructor_SucceedsWithServiceProvider()
{
// Arrange
var chatClient = new Mock<IChatClient>().Object;
var services = new Mock<IServiceProvider>().Object;
// Act
var agent = new HarnessAgent(chatClient, TestMaxContextWindowTokens, TestMaxOutputTokens, CreateAllDisabledOptions(), services: services);
// Assert
Assert.NotNull(agent);
}
/// <summary>
/// Verify that the constructor succeeds when both loggerFactory and serviceProvider are provided.
/// </summary>
[Fact]
public void Constructor_SucceedsWithLoggerFactoryAndServiceProvider()
{
// Arrange
var chatClient = new Mock<IChatClient>().Object;
var loggerFactory = new Mock<ILoggerFactory>().Object;
var services = new Mock<IServiceProvider>().Object;
// Act
var agent = new HarnessAgent(chatClient, TestMaxContextWindowTokens, TestMaxOutputTokens, CreateAllDisabledOptions(), loggerFactory, services);
// Assert
Assert.NotNull(agent);
}
/// <summary>
/// Verify that AsHarnessAgent extension method accepts loggerFactory and serviceProvider.
/// </summary>
[Fact]
public void AsHarnessAgent_SucceedsWithLoggerFactoryAndServiceProvider()
{
// Arrange
var chatClient = new Mock<IChatClient>().Object;
var loggerFactory = new Mock<ILoggerFactory>().Object;
var services = new Mock<IServiceProvider>().Object;
// Act
var agent = chatClient.AsHarnessAgent(TestMaxContextWindowTokens, TestMaxOutputTokens, CreateAllDisabledOptions(), loggerFactory, services);
// Assert
Assert.NotNull(agent);
}
/// <summary>
/// Verify that ILoggerFactory is threaded to downstream components by confirming CreateLogger is called.
/// </summary>
[Fact]
public void Constructor_LoggerFactoryIsUsedByDownstreamComponents()
{
// Arrange
var chatClient = new Mock<IChatClient>().Object;
var mockLoggerFactory = new Mock<ILoggerFactory>();
mockLoggerFactory
.Setup(lf => lf.CreateLogger(It.IsAny<string>()))
.Returns(new Mock<ILogger>().Object);
// Act — use options that leave CompactionProvider and AgentSkillsProvider enabled
var options = new HarnessAgentOptions
{
DisableToolApproval = true,
DisableOpenTelemetry = true,
DisableFileMemory = true,
DisableFileAccess = true,
DisableWebSearch = true,
DisableTodoProvider = true,
DisableAgentModeProvider = true,
};
var agent = new HarnessAgent(chatClient, TestMaxContextWindowTokens, TestMaxOutputTokens, options, mockLoggerFactory.Object);
// Assert — CreateLogger should have been called by one or more downstream components
Assert.NotNull(agent);
mockLoggerFactory.Verify(lf => lf.CreateLogger(It.IsAny<string>()), Times.AtLeastOnce());
}
/// <summary>
/// Verify that IServiceProvider is propagated through the agent pipeline by confirming
/// it is queried during agent construction.
/// </summary>
[Fact]
public void Constructor_ServiceProviderIsQueriedDuringBuild()
{
// Arrange
var chatClient = new Mock<IChatClient>().Object;
var mockServices = new Mock<IServiceProvider>();
mockServices
.Setup(sp => sp.GetService(It.IsAny<Type>()))
.Returns(null!);
// Act
var agent = new HarnessAgent(chatClient, TestMaxContextWindowTokens, TestMaxOutputTokens, CreateAllDisabledOptions(), services: mockServices.Object);
// Assert — the service provider should have been queried during pipeline construction
Assert.NotNull(agent);
mockServices.Verify(sp => sp.GetService(It.IsAny<Type>()), Times.AtLeastOnce());
}
#endregion
}
@@ -1,12 +0,0 @@
{
"profiles": {
"Microsoft.Agents.AI.Hosting.A2A.UnitTests": {
"commandName": "Project",
"launchBrowser": true,
"environmentVariables": {
"ASPNETCORE_ENVIRONMENT": "Development"
},
"applicationUrl": "https://localhost:52186;http://localhost:52187"
}
}
}
@@ -1,12 +0,0 @@
{
"profiles": {
"Microsoft.Agents.AI.Hosting.OpenAI.UnitTests": {
"commandName": "Project",
"launchBrowser": true,
"environmentVariables": {
"ASPNETCORE_ENVIRONMENT": "Development"
},
"applicationUrl": "https://localhost:60491;http://localhost:60492"
}
}
}
@@ -51,9 +51,8 @@ public sealed class AgentClassSkillTests
// Act & Assert — Content is cached
Assert.Same(await skill.GetContentAsync(), await skill.GetContentAsync());
// Act & Assert — Content includes parameter schema from typed script
Assert.Contains("parameters_schema", await skill.GetContentAsync());
Assert.Contains("value", await skill.GetContentAsync());
// Act & Assert — Content includes parameter schema from typed script (with preserved quotes)
Assert.Contains("\"value\"", await skill.GetContentAsync());
}
[Fact]
@@ -383,10 +382,9 @@ public sealed class AgentClassSkillTests
// Arrange
var skill = new AttributedFullSkill();
// Act & Assert — Content includes reflected resources and scripts
Assert.Contains("<resources>", await skill.GetContentAsync());
Assert.Contains("conversion-table", await skill.GetContentAsync());
Assert.Contains("<scripts>", await skill.GetContentAsync());
// Act & Assert — Content no longer includes resources in body; scripts are in script_schemas
Assert.DoesNotContain("<resources>", await skill.GetContentAsync());
Assert.Contains("<script_schemas>", await skill.GetContentAsync());
Assert.Contains("convert", await skill.GetContentAsync());
// Act & Assert — discovered members are cached
@@ -504,7 +502,7 @@ public sealed class AgentClassSkillTests
}
[Fact]
public async Task Content_IncludesDescription_ForReflectedResourcesAsync()
public async Task Content_DoesNotRenderResources_InBodyAsync()
{
// Arrange
var skill = new AttributedResourcePropertiesSkill();
@@ -512,8 +510,8 @@ public sealed class AgentClassSkillTests
// Act
var content = await skill.GetContentAsync();
// Assert — descriptions from [Description] attribute appear in synthesized content
Assert.Contains("Some important data.", content);
// Assert — resources are no longer rendered in body content
Assert.DoesNotContain("<resources>", content);
}
[Fact]
@@ -122,11 +122,10 @@ public sealed class AgentFileSkillScriptTests
// Assert — content starts with original and appends per-script entries
Assert.StartsWith("Original content", content);
Assert.Contains("<scripts>", content);
Assert.Contains("<script name=\"build\">", content);
Assert.Contains("<script name=\"deploy\">", content);
Assert.Contains("<parameters_schema>", content);
Assert.Contains("</scripts>", content);
Assert.Contains("<script_schemas>", content);
Assert.Contains("<schema script=\"build\">", content);
Assert.Contains("<schema script=\"deploy\">", content);
Assert.Contains("</script_schemas>", content);
}
[Fact]
@@ -149,7 +149,7 @@ public sealed class AgentInlineSkillTests
}
[Fact]
public async Task Content_IncludesResourcesAddedBeforeFirstAccessAsync()
public async Task Content_DoesNotIncludeResourcesInBodyAsync()
{
// Arrange
var skill = new AgentInlineSkill("my-skill", "A valid skill.", "Instructions.");
@@ -158,13 +158,12 @@ public sealed class AgentInlineSkillTests
// Act
var content = await skill.GetContentAsync();
// Assert
Assert.Contains("<resources>", content);
Assert.Contains("config", content);
// Assert — resources are no longer rendered in the body; they're accessed via GetResourceAsync
Assert.DoesNotContain("<resources>", content);
}
[Fact]
public async Task Content_IncludesDelegateResourcesAddedBeforeFirstAccessAsync()
public async Task Content_DoesNotIncludeDelegateResourcesInBodyAsync()
{
// Arrange
var skill = new AgentInlineSkill("my-skill", "A valid skill.", "Instructions.");
@@ -173,9 +172,8 @@ public sealed class AgentInlineSkillTests
// Act
var content = await skill.GetContentAsync();
// Assert
Assert.Contains("<resources>", content);
Assert.Contains("dynamic", content);
// Assert — resources are no longer rendered in the body
Assert.DoesNotContain("<resources>", content);
}
[Fact]
@@ -189,7 +187,7 @@ public sealed class AgentInlineSkillTests
var content = await skill.GetContentAsync();
// Assert
Assert.Contains("<scripts>", content);
Assert.Contains("<script_schemas>", content);
Assert.Contains("run", content);
}
@@ -209,7 +207,7 @@ public sealed class AgentInlineSkillTests
}
[Fact]
public async Task Content_IncludesResourcesAndScriptsAddedBeforeFirstAccessAsync()
public async Task Content_IncludesScriptSchemasAddedBeforeFirstAccessAsync()
{
// Arrange
var skill = new AgentInlineSkill("my-skill", "A valid skill.", "Instructions.");
@@ -220,9 +218,8 @@ public sealed class AgentInlineSkillTests
var content = await skill.GetContentAsync();
// Assert
Assert.Contains("<resources>", content);
Assert.Contains("r1", content);
Assert.Contains("<scripts>", content);
Assert.DoesNotContain("<resources>", content);
Assert.Contains("<script_schemas>", content);
Assert.Contains("s1", content);
}
@@ -236,8 +233,9 @@ public sealed class AgentInlineSkillTests
// Act
var content = await skill.GetContentAsync();
// Assert — JSON schema should be present and XML content chars escaped
Assert.Contains("parameters_schema", content);
// Assert — JSON schema should be present inside <schema> element (no extra wrapper) with preserved quotes
Assert.Contains("<schema script=\"search\">", content);
Assert.Contains("\"query\"", content);
Assert.DoesNotContain("<![CDATA[", content);
}
@@ -429,7 +427,7 @@ public sealed class AgentInlineSkillTests
// Assert
Assert.DoesNotContain("<resources>", content);
Assert.DoesNotContain("<scripts>", content);
Assert.DoesNotContain("<script_schemas>", content);
}
[Fact]
@@ -463,7 +461,7 @@ public sealed class AgentInlineSkillTests
}
[Fact]
public async Task Content_ScriptWithDescription_IncludesDescriptionAttributeAsync()
public async Task Content_ScriptWithDescription_DoesNotEmitDescriptionAttributeAsync()
{
// Arrange
var skill = new AgentInlineSkill("my-skill", "A valid skill.", "Instructions.");
@@ -472,8 +470,10 @@ public sealed class AgentInlineSkillTests
// Act
var content = await skill.GetContentAsync();
// Assert
Assert.Contains("description=\"Runs something.\"", content);
// Assert — description is no longer emitted in the script_schemas block;
// the block only contains parameter schemas for calling scripts.
Assert.Contains("<schema script=\"my-script\"", content);
Assert.DoesNotContain("description=\"Runs something.\"", content);
}
[Fact]
@@ -492,7 +492,7 @@ public sealed class AgentInlineSkillTests
}
[Fact]
public async Task Content_ResourceWithDescription_IncludesDescriptionAttributeAsync()
public async Task Content_ResourceWithDescription_NotRenderedInBodyAsync()
{
// Arrange
var skill = new AgentInlineSkill("my-skill", "A valid skill.", "Instructions.");
@@ -502,9 +502,10 @@ public sealed class AgentInlineSkillTests
// Act
var content = await skill.GetContentAsync();
// Assert
Assert.Contains("description=\"A described resource.\"", content);
Assert.DoesNotContain("no-desc\" description", content);
// Assert — resources are no longer rendered in the body
Assert.DoesNotContain("<resources>", content);
Assert.DoesNotContain("with-desc", content);
Assert.DoesNotContain("no-desc", content);
}
[Fact]
@@ -2,6 +2,7 @@
using System.Collections.Generic;
using System.Linq;
using System.Reflection;
using System.Threading;
using System.Threading.Tasks;
using Microsoft.Agents.AI.Workflows.Declarative.Events;
@@ -11,7 +12,9 @@ using Microsoft.Agents.AI.Workflows.Declarative.ObjectModel;
using Microsoft.Agents.AI.Workflows.Declarative.PowerFx;
using Microsoft.Agents.ObjectModel;
using Microsoft.Extensions.AI;
using Microsoft.PowerFx.Types;
using Moq;
using ApprovalSnapshot = Microsoft.Agents.AI.Workflows.Declarative.ObjectModel.InvokeMcpToolExecutor.ApprovalSnapshot;
namespace Microsoft.Agents.AI.Workflows.Declarative.UnitTests.ObjectModel;
@@ -842,6 +845,313 @@ public sealed class InvokeMcpToolExecutorTest(ITestOutputHelper output) : Workfl
#endregion
#region Approval Snapshot Security Tests
/// <summary>
/// Verifies that mutating the tool name variable after approval does not change
/// which tool is actually invoked. The originally-approved tool name must be used.
/// </summary>
[Fact]
public async Task InvokeMcpToolCaptureResponseUsesApprovedToolNameNotMutatedAsync()
{
// Arrange
const string ApprovedToolName = "safe_readonly_query";
const string MutatedToolName = "dangerous_admin_tool";
this.State.Set("TargetTool", FormulaValue.New(ApprovedToolName));
this.State.InitializeSystem();
this.State.Bind();
InvokeMcpTool model = this.CreateModelWithVariableToolName(
displayName: nameof(InvokeMcpToolCaptureResponseUsesApprovedToolNameNotMutatedAsync),
serverUrl: TestServerUrl,
variableName: "TargetTool");
string? capturedToolName = null;
Mock<IMcpToolHandler> mockProvider = new();
mockProvider.Setup(provider => provider.InvokeToolAsync(
It.IsAny<string>(),
It.IsAny<string?>(),
It.IsAny<string>(),
It.IsAny<IDictionary<string, object?>?>(),
It.IsAny<IDictionary<string, string>?>(),
It.IsAny<string?>(),
It.IsAny<CancellationToken>()))
.Callback<string, string?, string, IDictionary<string, object?>?, IDictionary<string, string>?, string?, CancellationToken>(
(_, _, toolName, _, _, _, _) => capturedToolName = toolName)
.ReturnsAsync(new McpServerToolResultContent("capture-call-id")
{
Outputs = [new TextContent("result")]
});
MockAgentProvider mockAgentProvider = new();
InvokeMcpToolExecutor action = new(model, mockProvider.Object, mockAgentProvider.Object, this.State);
// Act - trigger ExecuteAsync to store the approval snapshot
Mock<IWorkflowContext> mockContext = CreateMockWorkflowContext();
await action.HandleAsync(new ActionExecutorResult(action.Id), mockContext.Object, CancellationToken.None);
// Simulate parallel branch mutating state during the approval window
this.State.Set("TargetTool", FormulaValue.New(MutatedToolName));
this.State.Bind();
// User clicks approve (they saw "safe_readonly_query" in the approval UI)
McpServerToolCallContent toolCall = new(action.Id, ApprovedToolName, TestServerUrl);
ToolApprovalRequestContent approvalRequest = new(action.Id, toolCall);
ToolApprovalResponseContent approvalResponse = approvalRequest.CreateResponse(approved: true);
ExternalInputResponse response = new(new ChatMessage(ChatRole.User, [approvalResponse]));
// Resume after approval
await action.CaptureResponseAsync(mockContext.Object, response, CancellationToken.None);
// Assert - the originally-approved tool name must be used, not the mutated one
Assert.NotNull(capturedToolName);
Assert.Equal(ApprovedToolName, capturedToolName);
}
/// <summary>
/// Verifies that mutating an argument variable after approval does not change
/// the arguments actually passed to the MCP tool. The originally-approved arguments must be used.
/// </summary>
[Fact]
public async Task InvokeMcpToolCaptureResponseUsesApprovedArgumentsNotMutatedAsync()
{
// Arrange
const string ApprovedQuery = "SELECT * FROM users LIMIT 10";
const string MutatedQuery = "DROP TABLE users CASCADE; --";
this.State.Set("SqlQuery", FormulaValue.New(ApprovedQuery));
this.State.InitializeSystem();
this.State.Bind();
InvokeMcpTool model = this.CreateModelWithVariableArgument(
displayName: nameof(InvokeMcpToolCaptureResponseUsesApprovedArgumentsNotMutatedAsync),
serverUrl: TestServerUrl,
toolName: TestToolName,
argumentKey: "query",
variableName: "SqlQuery");
IDictionary<string, object?>? capturedArguments = null;
Mock<IMcpToolHandler> mockProvider = new();
mockProvider.Setup(provider => provider.InvokeToolAsync(
It.IsAny<string>(),
It.IsAny<string?>(),
It.IsAny<string>(),
It.IsAny<IDictionary<string, object?>?>(),
It.IsAny<IDictionary<string, string>?>(),
It.IsAny<string?>(),
It.IsAny<CancellationToken>()))
.Callback<string, string?, string, IDictionary<string, object?>?, IDictionary<string, string>?, string?, CancellationToken>(
(_, _, _, arguments, _, _, _) => capturedArguments = arguments)
.ReturnsAsync(new McpServerToolResultContent("capture-call-id")
{
Outputs = [new TextContent("result")]
});
MockAgentProvider mockAgentProvider = new();
InvokeMcpToolExecutor action = new(model, mockProvider.Object, mockAgentProvider.Object, this.State);
// Act - trigger ExecuteAsync to store the approval snapshot
Mock<IWorkflowContext> mockContext = CreateMockWorkflowContext();
await action.HandleAsync(new ActionExecutorResult(action.Id), mockContext.Object, CancellationToken.None);
// Simulate parallel branch mutating state during the approval window
this.State.Set("SqlQuery", FormulaValue.New(MutatedQuery));
this.State.Bind();
// User clicks approve
McpServerToolCallContent toolCall = new(action.Id, TestToolName, TestServerUrl);
ToolApprovalRequestContent approvalRequest = new(action.Id, toolCall);
ToolApprovalResponseContent approvalResponse = approvalRequest.CreateResponse(approved: true);
ExternalInputResponse response = new(new ChatMessage(ChatRole.User, [approvalResponse]));
// Resume after approval
await action.CaptureResponseAsync(mockContext.Object, response, CancellationToken.None);
// Assert - the originally-approved argument must be used, not the mutated one
Assert.NotNull(capturedArguments);
Assert.Equal(ApprovedQuery, capturedArguments["query"]?.ToString());
}
/// <summary>
/// Verifies that mutating the server URL variable after approval does not redirect
/// the MCP tool call to a different server. The originally-approved server URL must be used.
/// </summary>
[Fact]
public async Task InvokeMcpToolCaptureResponseUsesApprovedServerUrlNotMutatedAsync()
{
// Arrange
const string ApprovedServerUrl = "https://internal-mcp.corp";
const string MutatedServerUrl = "https://attacker.evil/steal";
this.State.Set("McpEndpoint", FormulaValue.New(ApprovedServerUrl));
this.State.InitializeSystem();
this.State.Bind();
InvokeMcpTool model = this.CreateModelWithVariableServerUrl(
displayName: nameof(InvokeMcpToolCaptureResponseUsesApprovedServerUrlNotMutatedAsync),
variableName: "McpEndpoint",
toolName: TestToolName);
string? capturedServerUrl = null;
Mock<IMcpToolHandler> mockProvider = new();
mockProvider.Setup(provider => provider.InvokeToolAsync(
It.IsAny<string>(),
It.IsAny<string?>(),
It.IsAny<string>(),
It.IsAny<IDictionary<string, object?>?>(),
It.IsAny<IDictionary<string, string>?>(),
It.IsAny<string?>(),
It.IsAny<CancellationToken>()))
.Callback<string, string?, string, IDictionary<string, object?>?, IDictionary<string, string>?, string?, CancellationToken>(
(serverUrl, _, _, _, _, _, _) => capturedServerUrl = serverUrl)
.ReturnsAsync(new McpServerToolResultContent("capture-call-id")
{
Outputs = [new TextContent("result")]
});
MockAgentProvider mockAgentProvider = new();
InvokeMcpToolExecutor action = new(model, mockProvider.Object, mockAgentProvider.Object, this.State);
// Act - trigger ExecuteAsync to store the approval snapshot
Mock<IWorkflowContext> mockContext = CreateMockWorkflowContext();
await action.HandleAsync(new ActionExecutorResult(action.Id), mockContext.Object, CancellationToken.None);
// Simulate parallel branch mutating state during the approval window
this.State.Set("McpEndpoint", FormulaValue.New(MutatedServerUrl));
this.State.Bind();
// User clicks approve
McpServerToolCallContent toolCall = new(action.Id, TestToolName, ApprovedServerUrl);
ToolApprovalRequestContent approvalRequest = new(action.Id, toolCall);
ToolApprovalResponseContent approvalResponse = approvalRequest.CreateResponse(approved: true);
ExternalInputResponse response = new(new ChatMessage(ChatRole.User, [approvalResponse]));
// Resume after approval
await action.CaptureResponseAsync(mockContext.Object, response, CancellationToken.None);
// Assert - the originally-approved server URL must be used, not the mutated one
Assert.NotNull(capturedServerUrl);
Assert.Equal(ApprovedServerUrl, capturedServerUrl);
}
/// <summary>
/// Verifies that the approval snapshot survives a checkpoint/restore cycle.
/// After restore, the originally-approved tool name must still be used even if state was mutated.
/// </summary>
[Fact]
public async Task InvokeMcpToolCaptureResponseUsesSnapshotAfterCheckpointRestoreAsync()
{
// Arrange
const string ApprovedToolName = "safe_readonly_query";
const string MutatedToolName = "dangerous_admin_tool";
this.State.Set("TargetTool", FormulaValue.New(ApprovedToolName));
this.State.InitializeSystem();
this.State.Bind();
InvokeMcpTool model = this.CreateModelWithVariableToolName(
displayName: nameof(InvokeMcpToolCaptureResponseUsesSnapshotAfterCheckpointRestoreAsync),
serverUrl: TestServerUrl,
variableName: "TargetTool");
string? capturedToolName = null;
Mock<IMcpToolHandler> mockProvider = new();
mockProvider.Setup(provider => provider.InvokeToolAsync(
It.IsAny<string>(),
It.IsAny<string?>(),
It.IsAny<string>(),
It.IsAny<IDictionary<string, object?>?>(),
It.IsAny<IDictionary<string, string>?>(),
It.IsAny<string?>(),
It.IsAny<CancellationToken>()))
.Callback<string, string?, string, IDictionary<string, object?>?, IDictionary<string, string>?, string?, CancellationToken>(
(_, _, toolName, _, _, _, _) => capturedToolName = toolName)
.ReturnsAsync(new McpServerToolResultContent("capture-call-id")
{
Outputs = [new TextContent("result")]
});
MockAgentProvider mockAgentProvider = new();
InvokeMcpToolExecutor action = new(model, mockProvider.Object, mockAgentProvider.Object, this.State);
// Act - trigger ExecuteAsync to store the approval snapshot
Mock<IWorkflowContext> mockContext = CreateMockWorkflowContextWithStateStore();
await action.HandleAsync(new ActionExecutorResult(action.Id), mockContext.Object, CancellationToken.None);
// Simulate checkpoint: persist to state store
await InvokeProtectedMethodAsync(action, "OnCheckpointingAsync", mockContext.Object, CancellationToken.None);
// Simulate restore on a "new" executor instance by clearing the in-memory field via reflection
// (In production, a new executor instance would be created with _approvalSnapshot == null)
typeof(InvokeMcpToolExecutor)
.GetField("_approvalSnapshot", BindingFlags.NonPublic | BindingFlags.Instance)!
.SetValue(action, null);
// Restore from state store
await InvokeProtectedMethodAsync(action, "OnCheckpointRestoredAsync", mockContext.Object, CancellationToken.None);
// Mutate state after restore (simulating parallel branch)
this.State.Set("TargetTool", FormulaValue.New(MutatedToolName));
this.State.Bind();
// User clicks approve
McpServerToolCallContent toolCall = new(action.Id, ApprovedToolName, TestServerUrl);
ToolApprovalRequestContent approvalRequest = new(action.Id, toolCall);
ToolApprovalResponseContent approvalResponse = approvalRequest.CreateResponse(approved: true);
ExternalInputResponse response = new(new ChatMessage(ChatRole.User, [approvalResponse]));
// Resume after approval
await action.CaptureResponseAsync(mockContext.Object, response, CancellationToken.None);
// Assert - the originally-approved tool name must be used, not the mutated one
Assert.NotNull(capturedToolName);
Assert.Equal(ApprovedToolName, capturedToolName);
}
private static Mock<IWorkflowContext> CreateMockWorkflowContext()
{
Mock<IWorkflowContext> mockContext = new();
mockContext.Setup(c => c.AddEventAsync(It.IsAny<WorkflowEvent>(), It.IsAny<CancellationToken>()))
.Returns(default(ValueTask));
mockContext.Setup(c => c.QueueStateUpdateAsync(It.IsAny<string>(), It.IsAny<object?>(), It.IsAny<string?>(), It.IsAny<CancellationToken>()))
.Returns(default(ValueTask));
mockContext.Setup(c => c.SendMessageAsync(It.IsAny<object>(), It.IsAny<string?>(), It.IsAny<CancellationToken>()))
.Returns(default(ValueTask));
return mockContext;
}
/// <summary>
/// Creates a mock workflow context that actually stores state values (for checkpoint/restore tests).
/// </summary>
private static Mock<IWorkflowContext> CreateMockWorkflowContextWithStateStore()
{
Dictionary<string, object?> stateStore = new();
Mock<IWorkflowContext> mockContext = new();
mockContext.Setup(c => c.AddEventAsync(It.IsAny<WorkflowEvent>(), It.IsAny<CancellationToken>()))
.Returns(default(ValueTask));
mockContext.Setup(c => c.QueueStateUpdateAsync(It.IsAny<string>(), It.IsAny<ApprovalSnapshot?>(), It.IsAny<string?>(), It.IsAny<CancellationToken>()))
.Callback<string, ApprovalSnapshot?, string?, CancellationToken>((key, value, _, _) => stateStore[key] = value)
.Returns(default(ValueTask));
mockContext.Setup(c => c.SendMessageAsync(It.IsAny<object>(), It.IsAny<string?>(), It.IsAny<CancellationToken>()))
.Returns(default(ValueTask));
mockContext.Setup(c => c.ReadStateAsync<ApprovalSnapshot>(It.IsAny<string>(), It.IsAny<string?>(), It.IsAny<CancellationToken>()))
.Returns<string, string?, CancellationToken>((key, _, _) =>
new ValueTask<ApprovalSnapshot?>(stateStore.TryGetValue(key, out object? val) ? val as ApprovalSnapshot : null));
mockContext.Setup(c => c.ReadStateKeysAsync(It.IsAny<string?>(), It.IsAny<CancellationToken>()))
.ReturnsAsync(new HashSet<string>());
return mockContext;
}
/// <summary>
/// Invokes a protected method on an executor via reflection (for testing checkpoint hooks).
/// </summary>
private static async ValueTask InvokeProtectedMethodAsync(InvokeMcpToolExecutor action, string methodName, IWorkflowContext context, CancellationToken cancellationToken)
{
MethodInfo method = typeof(InvokeMcpToolExecutor)
.GetMethod(methodName, BindingFlags.NonPublic | BindingFlags.Instance)!;
ValueTask result = (ValueTask)method.Invoke(action, [context, cancellationToken])!;
await result.ConfigureAwait(false);
}
#endregion
#region CompleteAsync Tests
[Fact]
@@ -951,6 +1261,50 @@ public sealed class InvokeMcpToolExecutorTest(ITestOutputHelper output) : Workfl
return AssignParent<InvokeMcpTool>(builder);
}
private InvokeMcpTool CreateModelWithVariableToolName(string displayName, string serverUrl, string variableName)
{
InvokeMcpTool.Builder builder = new()
{
Id = this.CreateActionId(),
DisplayName = this.FormatDisplayName(displayName),
ServerUrl = new StringExpression.Builder(StringExpression.Literal(serverUrl)),
ToolName = new StringExpression.Builder(
StringExpression.Variable(PropertyPath.TopicVariable(variableName))),
RequireApproval = new BoolExpression.Builder(BoolExpression.Literal(true)),
};
return AssignParent<InvokeMcpTool>(builder);
}
private InvokeMcpTool CreateModelWithVariableArgument(
string displayName, string serverUrl, string toolName, string argumentKey, string variableName)
{
InvokeMcpTool.Builder builder = new()
{
Id = this.CreateActionId(),
DisplayName = this.FormatDisplayName(displayName),
ServerUrl = new StringExpression.Builder(StringExpression.Literal(serverUrl)),
ToolName = new StringExpression.Builder(StringExpression.Literal(toolName)),
RequireApproval = new BoolExpression.Builder(BoolExpression.Literal(true)),
};
builder.Arguments.Add(argumentKey,
ValueExpression.Variable(PropertyPath.TopicVariable(variableName)));
return AssignParent<InvokeMcpTool>(builder);
}
private InvokeMcpTool CreateModelWithVariableServerUrl(string displayName, string variableName, string toolName)
{
InvokeMcpTool.Builder builder = new()
{
Id = this.CreateActionId(),
DisplayName = this.FormatDisplayName(displayName),
ServerUrl = new StringExpression.Builder(
StringExpression.Variable(PropertyPath.TopicVariable(variableName))),
ToolName = new StringExpression.Builder(StringExpression.Literal(toolName)),
RequireApproval = new BoolExpression.Builder(BoolExpression.Literal(true)),
};
return AssignParent<InvokeMcpTool>(builder);
}
#endregion
#region Mock MCP Tool Provider
+39 -1
View File
@@ -7,6 +7,43 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
## [Unreleased]
## [1.8.0] - 2026-06-04
### Added
- **agent-framework-core**: Add MCP-based skills discovery (`McpSkillsSource`) ([#6169](https://github.com/microsoft/agent-framework/pull/6169))
- **agent-framework-core**: Progressive tool exposure via `FunctionInvocationContext` ([#6233](https://github.com/microsoft/agent-framework/pull/6233))
- **agent-framework-core**: Add background agent support to harness agent ([#6155](https://github.com/microsoft/agent-framework/pull/6155))
- **agent-framework-core**: Add `AgentFileStore` and `FileAccessProvider` for file access operations ([#6099](https://github.com/microsoft/agent-framework/pull/6099))
- **agent-framework-core**: Coalesce code interpreter history chunks ([#5801](https://github.com/microsoft/agent-framework/pull/5801))
- **agent-framework-core**: Run sync tools off the event loop ([#5773](https://github.com/microsoft/agent-framework/pull/5773))
- **agent-framework-bedrock**: Implement native structured output support via Converse API ([#6052](https://github.com/microsoft/agent-framework/pull/6052))
- **agent-framework-foundry**: Add Foundry Adaptive Evals integration for rubric-generation ([#6101](https://github.com/microsoft/agent-framework/pull/6101))
- **agent-framework-foundry**: Add `timeout` parameter to `FoundryAgent` to fix `ConnectTimeout` on multi-turn conversations ([#6263](https://github.com/microsoft/agent-framework/pull/6263))
- **agent-framework-mistral**: Add Mistral AI embedding client package ([#5480](https://github.com/microsoft/agent-framework/pull/5480))
- **agent-framework-a2a**: Expose `supported_protocol_bindings` as configurable parameter ([#6098](https://github.com/microsoft/agent-framework/pull/6098))
- **agent-framework-a2a**: Set `message_id` on `AgentResponseUpdate` for message-bearing paths ([#6163](https://github.com/microsoft/agent-framework/pull/6163))
- **agent-framework-foundry-hosting**: Persist hosted MCP call/results as canonical `mcp_call` output ([#6070](https://github.com/microsoft/agent-framework/pull/6070))
### Changed
- **agent-framework-github-copilot**: [BREAKING] Upgrade `github-copilot-sdk` to v1.0.0 (stable) ([#6292](https://github.com/microsoft/agent-framework/pull/6292))
- **agent-framework-core**: [BREAKING — experimental] Refactor Skill API to async resource and script lookup ([#6135](https://github.com/microsoft/agent-framework/pull/6135))
- **agent-framework-github-copilot**: Promote to release candidate (`1.0.0rc1`)
- **agent-framework-declarative**: Promote to release candidate (`1.0.0rc1`) ([#6256](https://github.com/microsoft/agent-framework/pull/6256))
### Fixed
- **agent-framework-core**: Fix compaction message-id collisions and tool-loop summary persistence ([#6299](https://github.com/microsoft/agent-framework/pull/6299))
- **agent-framework-core**: Fix observability unsafe serialization of function-call arguments containing dataclass/framework objects ([#6026](https://github.com/microsoft/agent-framework/pull/6026))
- **agent-framework-core**: Consolidate MCP reliability fixes ([#6145](https://github.com/microsoft/agent-framework/pull/6145))
- **agent-framework-core**: Backfill chat span request model if unknown and response model is available ([#6160](https://github.com/microsoft/agent-framework/pull/6160))
- **agent-framework-anthropic**: Skip orphan anthropic thinking signatures ([#5784](https://github.com/microsoft/agent-framework/pull/5784))
- **agent-framework-foundry**: Fix `FoundryAgent` stripping model from `PromptAgent` requests ([#5526](https://github.com/microsoft/agent-framework/pull/5526))
- **agent-framework-foundry-hosting**: Fix toolbox consent flow in hosted agent ([#6249](https://github.com/microsoft/agent-framework/pull/6249))
- **agent-framework-foundry-hosting**: Drop hosted MCP calls when reasoning is stripped ([#6210](https://github.com/microsoft/agent-framework/pull/6210))
- **agent-framework-openai**: Fix OTLP HTTP base-endpoint losing `/v1/{signal}` auto-append ([#5913](https://github.com/microsoft/agent-framework/pull/5913))
- **agent-framework-openai**: Drop hosted MCP calls when reasoning is stripped ([#6210](https://github.com/microsoft/agent-framework/pull/6210))
- **agent-framework-orchestrations**: Fix spurious Magentic custom manager warning ([#6261](https://github.com/microsoft/agent-framework/pull/6261))
- **agent-framework-azurefunctions**: Fix integration test worker crashes on Py3.13 ([#4260](https://github.com/microsoft/agent-framework/pull/4260))
## [1.7.0] - 2026-05-28
### Added
@@ -1132,7 +1169,8 @@ Release candidate for **agent-framework-core** and **agent-framework-azure-ai**
For more information, see the [announcement blog post](https://devblogs.microsoft.com/foundry/introducing-microsoft-agent-framework-the-open-source-engine-for-agentic-ai-apps/).
[Unreleased]: https://github.com/microsoft/agent-framework/compare/python-1.7.0...HEAD
[Unreleased]: https://github.com/microsoft/agent-framework/compare/python-1.8.0...HEAD
[1.8.0]: https://github.com/microsoft/agent-framework/compare/python-1.7.0...python-1.8.0
[1.7.0]: https://github.com/microsoft/agent-framework/compare/python-1.6.0...python-1.7.0
[1.6.0]: https://github.com/microsoft/agent-framework/compare/python-1.5.0...python-1.6.0
[1.5.0]: https://github.com/microsoft/agent-framework/compare/python-1.4.0...python-1.5.0
+1 -1
View File
@@ -33,7 +33,7 @@ Status is grouped into these buckets:
| `agent-framework-foundry` | `python/packages/foundry` | `released` |
| `agent-framework-foundry-local` | `python/packages/foundry_local` | `beta` |
| `agent-framework-gemini` | `python/packages/gemini` | `alpha` |
| `agent-framework-github-copilot` | `python/packages/github_copilot` | `beta` |
| `agent-framework-github-copilot` | `python/packages/github_copilot` | `rc` |
| `agent-framework-hyperlight` | `python/packages/hyperlight` | `beta` |
| `agent-framework-lab` | `python/packages/lab` | `beta` |
| `agent-framework-mem0` | `python/packages/mem0` | `beta` |
+2 -2
View File
@@ -4,7 +4,7 @@ description = "A2A integration for Microsoft Agent Framework."
authors = [{ name = "Microsoft", email = "af-support@microsoft.com"}]
readme = "README.md"
requires-python = ">=3.10"
version = "1.0.0b260528"
version = "1.0.0b260604"
license-files = ["LICENSE"]
urls.homepage = "https://aka.ms/agent-framework"
urls.source = "https://github.com/microsoft/agent-framework/tree/main/python"
@@ -23,7 +23,7 @@ classifiers = [
"Typing :: Typed",
]
dependencies = [
"agent-framework-core>=1.7.0,<2",
"agent-framework-core>=1.8.0,<2",
"a2a-sdk>=1.0.0,<2",
]
@@ -803,6 +803,15 @@ class RawAnthropicClient(
}
a_content.append(mcp_result)
case "text_reasoning":
if content.text is None:
if (
content.protected_data
and a_content
and a_content[-1].get("type") == "thinking"
and "signature" not in a_content[-1]
):
a_content[-1]["signature"] = content.protected_data
continue
thinking_block: dict[str, Any] = {"type": "thinking", "thinking": content.text}
if content.protected_data:
thinking_block["signature"] = content.protected_data
+2 -2
View File
@@ -4,7 +4,7 @@ description = "Anthropic integration for Microsoft Agent Framework."
authors = [{ name = "Microsoft", email = "af-support@microsoft.com"}]
readme = "README.md"
requires-python = ">=3.10"
version = "1.0.0b260521"
version = "1.0.0b260604"
license-files = ["LICENSE"]
urls.homepage = "https://aka.ms/agent-framework"
urls.source = "https://github.com/microsoft/agent-framework/tree/main/python"
@@ -23,7 +23,7 @@ classifiers = [
"Typing :: Typed",
]
dependencies = [
"agent-framework-core>=1.6.0,<2",
"agent-framework-core>=1.8.0,<2",
"anthropic>=0.80.0,<0.80.1",
]
@@ -485,6 +485,48 @@ def test_prepare_message_for_anthropic_text_reasoning_with_signature(
assert result["content"][0]["signature"] == "sig_abc123"
def test_prepare_message_for_anthropic_attaches_signature_only_reasoning(
mock_anthropic_client: MagicMock,
) -> None:
client = create_test_anthropic_client(mock_anthropic_client)
message = Message(
role="assistant",
contents=[
Content.from_text_reasoning(text="Let me think about this..."),
Content.from_text_reasoning(text=None, protected_data="sig_abc123"),
],
)
result = client._prepare_message_for_anthropic(message)
assert result["content"] == [
{"type": "thinking", "thinking": "Let me think about this...", "signature": "sig_abc123"}
]
def test_prepare_message_for_anthropic_skips_orphan_signature_only_reasoning(
mock_anthropic_client: MagicMock,
) -> None:
client = create_test_anthropic_client(mock_anthropic_client)
message = Message(
role="assistant",
contents=[
Content.from_text_reasoning(text=None, protected_data="sig_abc123"),
Content.from_function_call(
call_id="call_123",
name="get_weather",
arguments={"location": "San Francisco"},
),
],
)
result = client._prepare_message_for_anthropic(message)
assert len(result["content"]) == 1
assert result["content"][0]["type"] == "tool_use"
assert result["content"][0]["id"] == "call_123"
def test_prepare_message_for_anthropic_mcp_server_tool_call(
mock_anthropic_client: MagicMock,
) -> None:
@@ -4,7 +4,7 @@ description = "Azure Functions integration for Microsoft Agent Framework."
authors = [{ name = "Microsoft", email = "af-support@microsoft.com"}]
readme = "README.md"
requires-python = ">=3.10"
version = "1.0.0b260521"
version = "1.0.0b260604"
license-files = ["LICENSE"]
urls.homepage = "https://aka.ms/agent-framework"
urls.source = "https://github.com/microsoft/agent-framework/tree/main/python"
@@ -22,8 +22,8 @@ classifiers = [
"Typing :: Typed",
]
dependencies = [
"agent-framework-core>=1.6.0,<2",
"agent-framework-durabletask>=1.0.0b260521,<2",
"agent-framework-core>=1.8.0,<2",
"agent-framework-durabletask>=1.0.0b260604,<2",
"azure-functions>=1.24.0,<2",
"azure-functions-durable>=1.3.1,<2",
]
@@ -4,6 +4,7 @@
from __future__ import annotations
import asyncio
import copy
import json
import logging
import sys
@@ -36,6 +37,7 @@ from agent_framework.observability import ChatTelemetryLayer
from boto3.session import Session as Boto3Session
from botocore.client import BaseClient
from botocore.config import Config as BotoConfig
from botocore.exceptions import ClientError
from pydantic import BaseModel
if sys.version_info >= (3, 13):
@@ -115,13 +117,20 @@ class BedrockChatOptions(ChatOptions[ResponseModelT], Generic[ResponseModelT], t
translates to ``toolConfig.tools``.
tool_choice: How the model should use tools,
translates to ``toolConfig.toolChoice``.
response_format: Structured output format. Accepts a Pydantic BaseModel
subclass or an OpenAI-style dict schema
(``{"json_schema": {"name": ..., "schema": ...}}``).
When provided, the Converse API request includes
``outputConfig.textFormat`` with the schema serialized as a JSON
string. ``ChatResponse.value`` will be populated with the parsed
model instance. Only supported on models that support
``outputConfig.textFormat``. Unsupported models raise a ValueError.
# Options not supported in Bedrock Converse API:
seed: Not supported.
frequency_penalty: Not supported.
presence_penalty: Not supported.
allow_multiple_tool_calls: Not supported (models handle parallel calls automatically).
response_format: Not directly supported (use model-specific prompting).
user: Not supported.
store: Not supported.
logit_bias: Not supported.
@@ -161,9 +170,6 @@ class BedrockChatOptions(ChatOptions[ResponseModelT], Generic[ResponseModelT], t
allow_multiple_tool_calls: None # type: ignore[misc]
"""Not supported. Bedrock models handle parallel tool calls automatically."""
response_format: None # type: ignore[misc]
"""Not directly supported. Use model-specific prompting for JSON output."""
user: None # type: ignore[misc]
"""Not supported in Bedrock Converse API."""
@@ -324,10 +330,28 @@ class BedrockChatClient(
return Boto3Session(**session_kwargs)
def _invoke_converse(self, request: Mapping[str, Any]) -> dict[str, Any]:
response = self._bedrock_client.converse(**request)
if not isinstance(response, Mapping):
raise ChatClientInvalidResponseException("Bedrock converse response must be a mapping.")
return response
try:
response = self._bedrock_client.converse(**request)
if not isinstance(response, Mapping):
raise ChatClientInvalidResponseException("Bedrock converse response must be a mapping.")
return response
except ClientError as e:
error_details = e.response.get("Error", {})
error_code = error_details.get("Code", "")
error_message = error_details.get("Message", "")
# "outputConfig" in error_message catches cases where Bedrock explicitly
# rejects the outputConfig field (unsupported model). Other ValidationExceptions
# (e.g. malformed schema shape, invalid property values) will not mention
# "outputConfig" and will bubble up as raw ClientError without being misdiagnosed.
if error_code == "ValidationException" and (
"outputconfig" in error_message.lower() or "outputconfig" in str(e).lower()
):
raise ValueError(
f"Model '{self.model}' does not support structured output via outputConfig.textFormat. "
"Check the model's Bedrock Converse outputConfig/textFormat support. "
f"AWS error Code: {error_code}. AWS error Message: {error_message}"
) from e
raise
@override
def _inner_get_response(
@@ -344,7 +368,7 @@ class BedrockChatClient(
# Streaming mode - simulate streaming by yielding a single update
async def _stream() -> AsyncIterable[ChatResponseUpdate]:
response = await asyncio.to_thread(self._invoke_converse, request)
parsed_response = self._process_converse_response(response)
parsed_response = self._process_converse_response(response, options)
contents = list(parsed_response.messages[0].contents if parsed_response.messages else [])
if parsed_response.usage_details:
contents.append(Content.from_usage(usage_details=parsed_response.usage_details)) # type: ignore[arg-type]
@@ -360,12 +384,12 @@ class BedrockChatClient(
raw_representation=parsed_response.raw_representation,
)
return self._build_response_stream(_stream())
return self._build_response_stream(_stream(), response_format=options.get("response_format"))
# Non-streaming mode
async def _get_response() -> ChatResponse:
raw_response = await asyncio.to_thread(self._invoke_converse, request)
return self._process_converse_response(raw_response)
return self._process_converse_response(raw_response, options)
return _get_response()
@@ -430,6 +454,9 @@ class BedrockChatClient(
if tool_config:
run_options["toolConfig"] = tool_config
if output_config := self._prepare_output_config(options.get("response_format")):
run_options["outputConfig"] = output_config
return run_options
def _prepare_bedrock_messages(
@@ -628,7 +655,9 @@ class BedrockChatClient(
def _generate_tool_call_id() -> str:
return f"tool-call-{uuid4().hex}"
def _process_converse_response(self, response: dict[str, Any]) -> ChatResponse:
def _process_converse_response(
self, response: dict[str, Any], options: Mapping[str, Any] | None = None
) -> ChatResponse:
"""Convert Bedrock Converse API response to ChatResponse."""
output = response.get("output") or {}
message = output.get("message") or {}
@@ -646,6 +675,7 @@ class BedrockChatClient(
usage_details=usage_details,
model=model,
finish_reason=finish_reason,
response_format=options.get("response_format") if options else None,
raw_representation=response,
)
@@ -728,6 +758,101 @@ class BedrockChatClient(
return None
return FINISH_REASON_MAP.get(reason.lower())
def _prepare_output_config(self, response_format: Any | None) -> dict[str, Any] | None:
"""Convert response_format into the AWS Bedrock outputConfig wire format.
Args:
response_format: A Pydantic model class or a dict schema, or None.
Returns:
A dict for the Converse API ``outputConfig`` parameter, or None if
response_format is not set.
"""
if response_format is None:
return None
if isinstance(response_format, Mapping):
if "json_schema" in response_format:
# Shape A — OpenAI-style wrapper
json_schema_config = response_format["json_schema"]
schema_src = json_schema_config.get("schema", {})
name = json_schema_config.get("name", "output_schema")
elif "schema" in response_format:
# Shape B — inner shape directly {"name": ..., "schema": ...}
schema_src = response_format["schema"]
name = response_format.get("name", "output_schema")
else:
# Shape C — assume entire dict is the raw schema
logger.warning(
"response_format dict has no 'json_schema' or 'schema' key; "
"treating entire dict as raw JSON schema."
)
schema_src = dict(response_format)
name = "output_schema"
if isinstance(schema_src, str):
schema_src = json.loads(schema_src)
schema = copy.deepcopy(schema_src)
else:
if not isinstance(response_format, type) or not issubclass(response_format, BaseModel):
raise TypeError("response_format must be None, a dict JSON schema, or a Pydantic BaseModel subclass.")
# response_format is a Pydantic model class
schema = response_format.model_json_schema()
name = response_format.__name__
self._set_additional_properties_false(schema)
json_schema: dict[str, Any] = {
"name": name,
"schema": json.dumps(schema),
}
description = getattr(response_format, "__doc__", None) if not isinstance(response_format, Mapping) else None
if description and isinstance(description, str) and description.strip():
json_schema["description"] = description.strip()
return {
"textFormat": {
"type": "json_schema",
"structure": {"jsonSchema": json_schema},
}
}
def _set_additional_properties_false(self, schema: dict[str, Any]) -> None:
"""Recursively set additionalProperties: false on all object types in a JSON schema.
AWS requires strict schema enforcement. This mirrors the approach used by
AnthropicChatClient._prepare_response_format().
Args:
schema: The JSON schema dict to modify in-place.
"""
visited: set[int] = set()
def walk(node: Any) -> None:
if isinstance(node, dict):
node_id = id(node)
if node_id in visited:
return
visited.add(node_id)
if node.get("type") == "object" or ("properties" in node and "type" not in node):
existing = node.get("additionalProperties")
if existing is None or existing is True:
node["additionalProperties"] = False
for value in node.values():
if isinstance(value, (dict, list)):
walk(value)
elif isinstance(node, list):
node_id = id(node)
if node_id in visited:
return
visited.add(node_id)
for item in node:
if isinstance(item, (dict, list)):
walk(item)
walk(schema)
def service_url(self) -> str:
"""Returns the service URL for the Bedrock runtime in the configured AWS region.
+2 -2
View File
@@ -4,7 +4,7 @@ description = "Amazon Bedrock integration for Microsoft Agent Framework."
authors = [{ name = "Microsoft", email = "af-support@microsoft.com"}]
readme = "README.md"
requires-python = ">=3.10"
version = "1.0.0b260521"
version = "1.0.0b260604"
license-files = ["LICENSE"]
urls.homepage = "https://aka.ms/agent-framework"
urls.source = "https://github.com/microsoft/agent-framework/tree/main/python"
@@ -23,7 +23,7 @@ classifiers = [
"Typing :: Typed",
]
dependencies = [
"agent-framework-core>=1.6.0,<2",
"agent-framework-core>=1.8.0,<2",
"boto3>=1.35.0,<2.0.0",
"botocore>=1.35.0,<2.0.0",
]
@@ -0,0 +1,383 @@
# Copyright (c) Microsoft. All rights reserved.
from __future__ import annotations
import copy
import json
from typing import Any
from unittest.mock import patch
import pytest
from agent_framework import Content, Message
from botocore.exceptions import ClientError
from pydantic import BaseModel
from agent_framework_bedrock import BedrockChatClient
# region Test models
class WeatherReport(BaseModel):
city: str
temperature: float
summary: str
class NestedAddress(BaseModel):
street: str
city: str
zip_code: str
class Person(BaseModel):
name: str
age: int
address: NestedAddress
# endregion
# region Helpers
class _StubBedrockRuntime:
"""Stub that records calls and returns a canned response."""
def __init__(self, response_text: str = "Bedrock says hi") -> None:
self.calls: list[dict[str, Any]] = []
self._response_text = response_text
def converse(self, **kwargs: Any) -> dict[str, Any]:
self.calls.append(kwargs)
return {
"modelId": kwargs["modelId"],
"responseId": "resp-structured",
"usage": {"inputTokens": 10, "outputTokens": 20, "totalTokens": 30},
"output": {
"completionReason": "end_turn",
"message": {
"id": "msg-structured",
"role": "assistant",
"content": [{"text": self._response_text}],
},
},
}
def _make_client(response_text: str = "Bedrock says hi") -> tuple[BedrockChatClient, _StubBedrockRuntime]:
stub = _StubBedrockRuntime(response_text)
client = BedrockChatClient(
model="us.anthropic.claude-haiku-4-5-v1:0",
region="us-east-1",
client=stub,
)
return client, stub
def _user_messages() -> list[Message]:
return [Message(role="user", contents=[Content.from_text(text="Give me a weather report")])]
# endregion
# region Tests
def test_prepare_output_config_correct_wire_shape() -> None:
"""_prepare_output_config(WeatherReport) must produce the correct
textFormat → structure → jsonSchema shape with type: 'json_schema'."""
client, _ = _make_client()
output_config = client._prepare_output_config(WeatherReport)
assert output_config is not None
text_format = output_config["textFormat"]
assert text_format["type"] == "json_schema"
assert "structure" in text_format
json_schema = text_format["structure"]["jsonSchema"]
assert json_schema["name"] == "WeatherReport"
assert "schema" in json_schema
def test_prepare_output_config_schema_is_json_string() -> None:
"""The schema value inside jsonSchema must be a JSON string, not a dict."""
client, _ = _make_client()
output_config = client._prepare_output_config(WeatherReport)
assert output_config is not None
schema_value = output_config["textFormat"]["structure"]["jsonSchema"]["schema"]
assert isinstance(schema_value, str), f"Expected str, got {type(schema_value)}"
# Verify it's valid JSON
parsed = json.loads(schema_value)
assert isinstance(parsed, dict)
assert parsed["type"] == "object"
def test_additional_properties_false_set_recursively() -> None:
"""additionalProperties: false must be set on all nested object types."""
client, _ = _make_client()
output_config = client._prepare_output_config(Person)
assert output_config is not None
schema_str = output_config["textFormat"]["structure"]["jsonSchema"]["schema"]
schema = json.loads(schema_str)
# Top-level object
assert schema.get("additionalProperties") is False
# Check $defs for NestedAddress
defs = schema.get("$defs", {})
assert "NestedAddress" in defs, "Expected NestedAddress to be present in $defs"
assert defs["NestedAddress"].get("additionalProperties") is False, (
"Expected additionalProperties=False on nested NestedAddress schema"
)
def test_no_output_config_when_response_format_none() -> None:
"""When response_format is None, no outputConfig key should appear in the request."""
client, stub = _make_client()
messages = _user_messages()
request = client._prepare_options(messages, {"max_tokens": 100})
assert "outputConfig" not in request, (
f"outputConfig should not be present when response_format is None, got: {request.get('outputConfig')}"
)
async def test_chat_response_value_populated() -> None:
"""After a mocked response with response_format, .value should be a populated Pydantic model."""
json_response = json.dumps({"city": "Seattle", "temperature": 72.5, "summary": "Sunny and warm"})
client, stub = _make_client(response_text=json_response)
messages = _user_messages()
response = await client.get_response(
messages=messages,
options={"max_tokens": 100, "response_format": WeatherReport},
)
assert response.text == json_response
assert response.value is not None
assert isinstance(response.value, WeatherReport)
assert response.value.city == "Seattle"
assert response.value.temperature == 72.5
assert response.value.summary == "Sunny and warm"
# Verify outputConfig was sent to the API
assert len(stub.calls) == 1
api_request = stub.calls[0]
assert "outputConfig" in api_request
assert api_request["outputConfig"]["textFormat"]["type"] == "json_schema"
def test_dict_schema_response_format() -> None:
"""_prepare_output_config should work when response_format is a dict, not just a Pydantic class."""
client, _ = _make_client()
dict_schema = {
"json_schema": {
"name": "weather_output",
"schema": {
"type": "object",
"properties": {
"city": {"type": "string"},
"temp": {"type": "number"},
},
},
}
}
output_config = client._prepare_output_config(dict_schema)
assert output_config is not None
json_schema = output_config["textFormat"]["structure"]["jsonSchema"]
assert json_schema["name"] == "weather_output"
schema_parsed = json.loads(json_schema["schema"])
assert schema_parsed["type"] == "object"
assert "city" in schema_parsed["properties"]
def test_prepare_output_config_none_returns_none() -> None:
"""_prepare_output_config(None) must return None."""
client, _ = _make_client()
result = client._prepare_output_config(None)
assert result is None
async def test_chat_response_value_populated_streaming() -> None:
"""In streaming mode, .value should also be populated on the final response."""
json_response = json.dumps({"city": "Portland", "temperature": 68.0, "summary": "Cloudy"})
client, stub = _make_client(response_text=json_response)
messages = _user_messages()
stream = client.get_response(
messages=messages,
stream=True,
options={"max_tokens": 100, "response_format": WeatherReport},
)
# Consume stream and get final response
async for _ in stream:
pass
response = await stream.get_final_response()
assert response.value is not None
assert isinstance(response.value, WeatherReport)
assert response.value.city == "Portland"
# Verify outputConfig was sent
assert len(stub.calls) == 1
assert "outputConfig" in stub.calls[0]
async def test_unsupported_model_validation_exception() -> None:
"""When a model doesn't support outputConfig, a clear error should be raised."""
class _FailingStubBedrockRuntime:
def converse(self, **kwargs: Any) -> dict[str, Any]:
# Simulate botocore ClientError for ValidationException
error_response = {"Error": {"Code": "ValidationException", "Message": "Invalid field outputConfig"}}
raise ClientError(error_response, "Converse")
client = BedrockChatClient(
model="us.anthropic.claude-v2",
region="us-east-1",
client=_FailingStubBedrockRuntime(),
)
with pytest.raises(ValueError) as exc:
await client.get_response(
messages=_user_messages(),
options={"response_format": WeatherReport},
)
assert "does not support structured output via outputConfig.textFormat" in str(exc.value)
assert "Check the model's Bedrock Converse outputConfig/textFormat support." in str(exc.value)
def test_invalid_response_format_type_raises() -> None:
"""Non-dict, non-BaseModel response_format should raise TypeError."""
client, _ = _make_client()
with pytest.raises(TypeError, match="Pydantic BaseModel subclass"):
client._prepare_output_config("not_a_valid_format")
def test_mapping_response_format_accepted() -> None:
"""A non-dict Mapping response_format must be accepted and produce
correct outputConfig, not raise TypeError."""
from collections.abc import MutableMapping
class _WrappedMapping(MutableMapping):
def __init__(self, data):
self._data = dict(data)
def __getitem__(self, key):
return self._data[key]
def __setitem__(self, key, value):
self._data[key] = value
def __delitem__(self, key):
del self._data[key]
def __iter__(self):
return iter(self._data)
def __len__(self):
return len(self._data)
client, _ = _make_client()
mapping_format = _WrappedMapping({
"json_schema": {
"name": "test_output",
"schema": {
"type": "object",
"properties": {"result": {"type": "string"}},
},
}
})
output_config = client._prepare_output_config(mapping_format)
assert output_config is not None
json_schema = output_config["textFormat"]["structure"]["jsonSchema"]
assert json_schema["name"] == "test_output"
schema = json.loads(json_schema["schema"])
assert schema.get("additionalProperties") is False
def test_shape_b_dict_schema_wire_format() -> None:
"""Dict response_format in Shape B (inner shape directly) should
produce correct outputConfig."""
client, _ = _make_client()
response_format = {
"name": "weather_output",
"schema": {
"type": "object",
"properties": {
"city": {"type": "string"},
"temperature": {"type": "number"},
},
},
}
output_config = client._prepare_output_config(response_format)
assert output_config is not None
text_format = output_config["textFormat"]
assert text_format["type"] == "json_schema"
json_schema = text_format["structure"]["jsonSchema"]
assert json_schema["name"] == "weather_output"
schema = json.loads(json_schema["schema"])
assert schema.get("additionalProperties") is False
def test_dict_schema_not_mutated() -> None:
"""Caller's dict schema must not be mutated by _prepare_output_config."""
client, _ = _make_client()
original_schema = {
"json_schema": {
"name": "test",
"schema": {
"type": "object",
"properties": {"a": {"type": "string"}},
},
}
}
snapshot = copy.deepcopy(original_schema)
client._prepare_output_config(original_schema)
assert original_schema == snapshot, "Original dict schema was mutated"
async def test_non_outputconfig_validation_exception_propagates() -> None:
"""ValidationException unrelated to outputConfig must propagate
as raw ClientError, not be caught and reclassified."""
client, _ = _make_client()
error_response = {
"Error": {
"Code": "ValidationException",
"Message": "Invalid message format",
}
}
with (
patch.object(
client,
"_bedrock_client",
**{"converse.side_effect": ClientError(error_response, "Converse")},
),
pytest.raises(ClientError),
):
await client.get_response(
messages=_user_messages(),
options={"max_tokens": 100},
)
# endregion
+1 -1
View File
@@ -56,7 +56,7 @@ agent_framework/
- **`AgentMiddleware`** - Intercepts agent `run()` calls
- **`ChatMiddleware`** - Intercepts chat client `get_response()` calls
- **`FunctionMiddleware`** - Intercepts function/tool invocations
- **`AgentContext`** / **`ChatContext`** / **`FunctionInvocationContext`** - Context objects passed through middleware
- **`AgentContext`** / **`ChatContext`** / **`FunctionInvocationContext`** - Context objects passed through middleware. A tool can declare a `FunctionInvocationContext` parameter to receive it; `context.tools` is the live, mutable tools list for the run, and `context.add_tools(...)` / `context.remove_tools(...)` enable progressive tool exposure (changes apply on the next function-calling iteration).
### Sessions (`_sessions.py`)
@@ -71,6 +71,7 @@ from ._evaluation import (
Evaluator,
ExpectedToolCall,
LocalEvaluator,
RubricScore,
evaluate_agent,
evaluate_workflow,
evaluator,
@@ -167,6 +168,9 @@ from ._skills import (
InlineSkillResource,
InlineSkillScript,
InMemorySkillsSource,
MCPSkill,
MCPSkillResource,
MCPSkillsSource,
Skill,
SkillFrontmatter,
SkillResource,
@@ -443,6 +447,9 @@ __all__ = [
"MCPStdioTool",
"MCPStreamableHTTPTool",
"MCPWebsocketTool",
"MCPSkill",
"MCPSkillResource",
"MCPSkillsSource",
"MemoryContextProvider",
"MemoryFileStore",
"MemoryIndexEntry",
@@ -460,6 +467,7 @@ __all__ = [
"ResponseStream",
"Role",
"RoleLiteral",
"RubricScore",
"RunContext",
"Runner",
"RunnerContext",
@@ -380,8 +380,15 @@ class BaseChatClient(SerializationMixin, ABC, Generic[OptionsCoT]):
return prepared_messages
from ._compaction import apply_compaction
# Compact the caller's list in place when possible. A compaction operation has
# two halves: exclusion flags (mutated on shared Message objects) and inserted
# summary messages. Operating on the original list keeps both halves on the list
# the function-invocation tool loop reuses across iterations; otherwise inserted
# summaries would be lost on a throwaway copy while exclusions persisted, silently
# dropping older groups (issue #4991).
working_messages = messages if isinstance(messages, list) else prepared_messages
return await apply_compaction(
prepared_messages,
working_messages,
strategy=compaction_strategy,
tokenizer=tokenizer,
)
@@ -4,7 +4,7 @@ from __future__ import annotations
import json
import logging
from collections.abc import Mapping, Sequence
from collections.abc import Iterable, Mapping, Sequence
from typing import (
TYPE_CHECKING,
Any,
@@ -92,10 +92,23 @@ def _is_reasoning_only_assistant(message: Message) -> bool:
return all(content.type == "text_reasoning" for content in message.contents)
def _ensure_message_ids(messages: list[Message]) -> None:
def _ensure_message_ids(
messages: list[Message], *, id_offset: int = 0, reserved_ids: Iterable[str] | None = None
) -> None:
existing_ids: set[str] = set(reserved_ids) if reserved_ids is not None else set()
existing_ids.update(message.message_id for message in messages if message.message_id)
for index, message in enumerate(messages):
if not message.message_id:
message.message_id = f"msg_{index}"
if message.message_id:
continue
candidate = f"msg_{id_offset + index}"
if candidate in existing_ids:
counter = id_offset + len(messages)
candidate = f"msg_{counter}"
while candidate in existing_ids:
counter += 1
candidate = f"msg_{counter}"
message.message_id = candidate
existing_ids.add(candidate)
def _group_id_for(message: Message, group_index: int) -> str:
@@ -104,14 +117,27 @@ def _group_id_for(message: Message, group_index: int) -> str:
return f"group_index_{group_index}"
def group_messages(messages: list[Message]) -> list[dict[str, Any]]:
def group_messages(
messages: list[Message], *, id_offset: int = 0, reserved_ids: Iterable[str] | None = None
) -> list[dict[str, Any]]:
"""Compute group spans and metadata for annotation.
Args:
messages: The messages (or a slice of them) to group.
Keyword Args:
id_offset: Absolute starting index used when auto-assigning ``message_id``
values, so incremental annotation of a list slice produces ids that
stay unique across the full list.
reserved_ids: Message ids that already exist outside ``messages`` (for
example in a preserved prefix). Auto-assigned ids are guaranteed not
to collide with these, preventing duplicate ids across the full list.
Returns:
Ordered list of lightweight span dicts with keys:
``group_id``, ``kind``, ``start_index``, ``end_index``, ``has_reasoning``.
"""
_ensure_message_ids(messages)
_ensure_message_ids(messages, id_offset=id_offset, reserved_ids=reserved_ids)
spans: list[dict[str, Any]] = []
i = 0
group_index = 0
@@ -439,7 +465,8 @@ def annotate_message_groups(
if previous_group_index is not None:
group_index_offset = previous_group_index + 1
spans = group_messages(messages[start_index:])
reserved_ids = {message.message_id for message in messages[:start_index] if message.message_id}
spans = group_messages(messages[start_index:], id_offset=start_index, reserved_ids=reserved_ids)
for span_index, span in enumerate(spans):
group_id = str(span["group_id"])
kind = _coerce_group_kind(span["kind"])
@@ -311,12 +311,15 @@ class EvalScoreResult:
score: Numeric score from the evaluator.
passed: Whether the item passed this evaluator's threshold.
sample: Optional raw evaluator output (rationale, metadata).
dimensions: Per-dimension scores when this evaluator is a rubric
evaluator. ``None`` for non-rubric (e.g. built-in) evaluators.
"""
name: str
score: float
passed: bool | None = None
sample: dict[str, Any] | None = None
dimensions: list[RubricScore] | None = None
@experimental(feature_id=ExperimentalFeature.EVALS)
@@ -496,6 +499,179 @@ class EvalResults:
detail += f" Errored items: {', '.join(summaries)}."
raise EvalNotPassedError(detail)
def assert_score_at_least(
self,
min_score: float,
*,
evaluator: str | None = None,
msg: str | None = None,
) -> None:
"""Assert every item's score (optionally filtered by evaluator) is ``>= min_score``.
Designed for CI gates on generated rubric evaluators (e.g.
``results.assert_score_at_least(0.80)``). Includes any
sub-results from workflow evaluations.
Args:
min_score: Minimum acceptable score (inclusive).
evaluator: When set, only check scores from the evaluator
whose ``EvalScoreResult.name`` matches.
msg: Optional custom failure message.
Raises:
EvalNotPassedError: When any matching score is below the threshold.
"""
offenders: list[str] = []
def _check(results: EvalResults) -> None:
for item in results.items:
for score in item.scores:
if evaluator is not None and score.name != evaluator:
continue
if score.score < min_score:
offenders.append(f"{item.item_id}/{score.name}={score.score:.3f}")
for sub in results.sub_results.values():
_check(sub)
_check(self)
if offenders:
detail = msg or (
f"{len(offenders)} score(s) below threshold {min_score}"
f"{' for ' + evaluator if evaluator else ''}: {', '.join(offenders[:5])}"
+ (f" (+{len(offenders) - 5} more)" if len(offenders) > 5 else "")
)
raise EvalNotPassedError(detail)
def assert_dimension_score_at_least(
self,
dimension_id: str,
min_score: float,
*,
evaluator: str | None = None,
require_applicable: bool = False,
msg: str | None = None,
) -> None:
"""Assert every item's score for a rubric *dimension* is ``>= min_score``.
Walks ``EvalScoreResult.dimensions`` looking for the named
dimension across all items (and sub-results). Non-applicable
dimensions are skipped by default; pass
``require_applicable=True`` to fail when no applicable score is
produced.
Args:
dimension_id: Dimension id (matches the rubric definition).
min_score: Minimum acceptable dimension score (inclusive).
evaluator: When set, only consider scores from the evaluator
whose ``EvalScoreResult.name`` matches.
require_applicable: When ``True``, missing or non-applicable
dimension scores raise. Defaults to ``False`` (skip).
msg: Optional custom failure message.
Raises:
EvalNotPassedError: When the dimension fails the threshold.
"""
offenders: list[str] = []
missing_items: list[str] = []
def _check(results: EvalResults) -> None:
for item in results.items:
found_applicable = False
for score in item.scores:
if evaluator is not None and score.name != evaluator:
continue
if not score.dimensions:
continue
for rs in score.dimensions:
if rs.id != dimension_id:
continue
if not rs.applicable:
continue
found_applicable = True
if rs.score is None or rs.score < min_score:
offenders.append(
f"{item.item_id}/{score.name}/{dimension_id}="
f"{rs.score if rs.score is not None else 'None'}"
)
if require_applicable and not found_applicable:
missing_items.append(item.item_id)
for sub in results.sub_results.values():
_check(sub)
_check(self)
problems: list[str] = []
if offenders:
problems.append(
f"{len(offenders)} dimension score(s) for '{dimension_id}' below {min_score}: "
f"{', '.join(offenders[:5])}" + (f" (+{len(offenders) - 5} more)" if len(offenders) > 5 else "")
)
if missing_items:
problems.append(
f"Dimension '{dimension_id}' not applicable on {len(missing_items)} item(s): "
f"{', '.join(missing_items[:5])}"
)
if problems:
raise EvalNotPassedError(msg or "; ".join(problems))
def assert_no_failed_items(self, msg: str | None = None) -> None:
"""Assert no item ended in ``fail`` or ``error`` status.
Includes any sub-results from workflow evaluations.
Args:
msg: Optional custom failure message.
Raises:
EvalNotPassedError: When any item failed or errored.
"""
bad: list[str] = []
def _check(results: EvalResults) -> None:
for item in results.items:
if item.is_failed or item.is_error:
bad.append(f"{item.item_id}:{item.status}")
for sub in results.sub_results.values():
_check(sub)
_check(self)
if bad:
detail = msg or (
f"{len(bad)} item(s) failed or errored: {', '.join(bad[:5])}"
+ (f" (+{len(bad) - 5} more)" if len(bad) > 5 else "")
)
raise EvalNotPassedError(detail)
# endregion
# region Generated rubric evaluators
@experimental(feature_id=ExperimentalFeature.EVALS)
@dataclass(frozen=True)
class RubricScore:
"""A single dimension's score from a rubric-based evaluator run.
Rubric evaluators emit one ``RubricScore`` per dimension per item.
Attached to :class:`EvalScoreResult` as a typed view of the raw
``properties.rubric_scores`` payload returned by providers such as
Foundry's generated rubric evaluators.
Attributes:
id: Dimension id (matches the rubric definition).
score: Numeric score, or ``None`` when the dimension was marked
non-applicable for this item.
applicable: Whether the dimension applied to this item.
weight: Dimension weight (mirrors the rubric definition).
reason: Short rationale produced by the evaluator.
"""
id: str
score: int | None
applicable: bool
weight: int
reason: str
# endregion
@@ -58,6 +58,8 @@ class ExperimentalFeature(str, Enum):
FOUNDRY_PREVIEW_TOOLS = "FOUNDRY_PREVIEW_TOOLS"
FUNCTIONAL_WORKFLOWS = "FUNCTIONAL_WORKFLOWS"
HARNESS = "HARNESS"
MCP_SKILLS = "MCP_SKILLS"
PROGRESSIVE_TOOLS = "PROGRESSIVE_TOOLS"
SKILLS = "SKILLS"
TO_PROMPT_AGENT = "TO_PROMPT_AGENT"
@@ -349,6 +349,8 @@ class BackgroundAgentsProvider(ContextProvider):
_save_provider_state(session, provider_state, source_id=source_id)
return f"Background task {task_id} started on agent '{agent_name}'."
background_agents_start_task._invoke_sync_on_event_loop = True # pyright: ignore[reportPrivateUsage]
@tool(name="background_agents_wait_for_first_completion", approval_mode="never_require")
async def background_agents_wait_for_first_completion(task_ids: list[int]) -> str:
"""Block until the first of the specified background tasks completes. Returns the completed task's ID."""
@@ -471,6 +473,8 @@ class BackgroundAgentsProvider(ContextProvider):
_save_provider_state(session, provider_state, source_id=source_id)
return f"Task {task_id} continued with new input."
background_agents_continue_task._invoke_sync_on_event_loop = True # pyright: ignore[reportPrivateUsage]
@tool(name="background_agents_clear_completed_task", approval_mode="never_require")
def background_agents_clear_completed_task(task_id: int) -> str:
"""Remove a completed or failed task and release its session to free memory."""
@@ -11,6 +11,7 @@ from enum import Enum
from typing import TYPE_CHECKING, Any, Generic, Literal, TypeAlias, cast, overload
from ._clients import SupportsChatGetResponse
from ._feature_stage import ExperimentalFeature, experimental
from ._types import (
AgentResponse,
AgentResponseUpdate,
@@ -214,6 +215,12 @@ class FunctionInvocationContext:
result: Function execution result. Can be observed after calling ``call_next()``
to see the actual execution result or can be set to override the execution result.
kwargs: Additional runtime keyword arguments forwarded to the function invocation.
tools: The live, mutable list of tools available to the model for the current
agent run, or ``None`` when the function is invoked outside of a
function-calling loop (for example via ``FunctionTool.invoke`` directly).
Tools can add or remove tools during execution using :meth:`add_tools`
and :meth:`remove_tools` (progressive tool exposure). Mutations take
effect on the **next** model iteration, not the in-flight batch.
Examples:
.. code-block:: python
@@ -232,6 +239,18 @@ class FunctionInvocationContext:
# Continue execution
await call_next()
Progressive tool exposure from inside a tool:
.. code-block:: python
from agent_framework import FunctionInvocationContext, tool
@tool(approval_mode="never_require")
def load_math_tools(ctx: FunctionInvocationContext) -> str:
ctx.add_tools([factorial, fibonacci])
return "Math tools are now available."
"""
def __init__(
@@ -242,6 +261,7 @@ class FunctionInvocationContext:
metadata: Mapping[str, Any] | None = None,
result: Any = None,
kwargs: Mapping[str, Any] | None = None,
tools: list[ToolTypes] | None = None,
) -> None:
"""Initialize the FunctionInvocationContext.
@@ -252,6 +272,9 @@ class FunctionInvocationContext:
metadata: Metadata dictionary for sharing data between function middleware.
result: Function execution result.
kwargs: Additional runtime keyword arguments forwarded to the function invocation.
tools: The live, mutable list of tools for the current agent run. When provided,
this is the same list object the model sees on the next iteration, so
appending or removing tools changes the model's available tools.
"""
self.function = function
self.arguments = arguments
@@ -259,6 +282,96 @@ class FunctionInvocationContext:
self.metadata: dict[str, Any] = dict(metadata) if metadata is not None else {}
self.result = result
self.kwargs: dict[str, Any] = dict(kwargs) if kwargs is not None else {}
self.tools = tools
@experimental(feature_id=ExperimentalFeature.PROGRESSIVE_TOOLS)
def add_tools(
self,
tools: ToolTypes | Callable[..., Any] | Sequence[ToolTypes | Callable[..., Any]],
) -> None:
"""Add one or more tools to the current agent run (progressive tool exposure).
Callable inputs are converted to :class:`FunctionTool`, and tool collections are
flattened, using the same normalization as the rest of the framework. Added tools
become available to the model on the **next** iteration of the function-calling
loop; they do not affect tool calls already requested in the in-flight batch.
Adding a tool whose name already exists is a no-op when it is the same object, and
raises ``ValueError`` when it is a different object with a duplicate name.
Args:
tools: A single tool/callable or a sequence of tools/callables to add.
Raises:
RuntimeError: If the context has no live tools list (for example when the
function is invoked outside of a function-calling loop).
ValueError: If a different tool with a duplicate name is added.
"""
from ._tools import _append_unique_tools, normalize_tools # type: ignore[reportPrivateUsage]
if self.tools is None:
raise RuntimeError(
"Cannot add tools: this FunctionInvocationContext is not bound to a live "
"agent run. add_tools is only available for functions invoked within an "
"agent's function-calling loop."
)
# Validate the whole batch against a throwaway copy first, so a duplicate-name
# clash partway through the batch raises before the live tool list is mutated
# (all-or-nothing semantics).
merged = _append_unique_tools(list(self.tools), normalize_tools(tools))
self.tools[:] = merged
@experimental(feature_id=ExperimentalFeature.PROGRESSIVE_TOOLS)
def remove_tools(
self,
tools: ToolTypes | Callable[..., Any] | Sequence[ToolTypes | Callable[..., Any]] | str | Sequence[str],
) -> None:
"""Remove one or more tools from the current agent run (progressive tool exposure).
Tools may be specified by name, by tool object, or by the original callable. Names
that are not currently present are ignored. Removals take effect on the **next**
iteration of the function-calling loop; tool calls already requested in the
in-flight batch still execute.
Args:
tools: A tool name, tool/callable, or a sequence of any of these to remove.
Raises:
RuntimeError: If the context has no live tools list (for example when the
function is invoked outside of a function-calling loop).
"""
from ._tools import _get_tool_name, normalize_tools # type: ignore[reportPrivateUsage]
if self.tools is None:
raise RuntimeError(
"Cannot remove tools: this FunctionInvocationContext is not bound to a live "
"agent run. remove_tools is only available for functions invoked within an "
"agent's function-calling loop."
)
names_to_remove: set[str] = set()
raw_items: list[Any]
if isinstance(tools, str):
raw_items = [tools]
elif isinstance(tools, Sequence) and not isinstance(tools, (bytes, bytearray)):
raw_items = list(cast("Sequence[Any]", tools))
else:
raw_items = [tools]
for item in raw_items:
if isinstance(item, str):
names_to_remove.add(item)
continue
for normalized in normalize_tools(item):
if name := _get_tool_name(normalized): # type: ignore[reportPrivateUsage]
names_to_remove.add(name)
if not names_to_remove:
return
self.tools[:] = [
tool
for tool in self.tools
if _get_tool_name(tool) not in names_to_remove # type: ignore[reportPrivateUsage]
]
class ChatContext:
@@ -7,6 +7,8 @@ import json
import logging
import re
from collections.abc import Mapping, MutableMapping
from dataclasses import asdict, is_dataclass
from datetime import date, datetime
from typing import Any, ClassVar, Protocol, TypeVar, runtime_checkable
logger = logging.getLogger("agent_framework")
@@ -614,3 +616,46 @@ class SerializationMixin:
# Fallback and default
# Convert class name to snake_case
return _CAMEL_TO_SNAKE_PATTERN.sub("_", cls.__name__).lower()
def make_json_safe(obj: Any) -> Any:
"""Recursively convert an object to a JSON-serializable form.
Handles dataclasses, Pydantic models, objects with ``to_dict``/``dict``/``__dict__``,
datetimes, lists, dicts, and primitives. Falls back to ``str()`` for any remaining
non-serializable value so that ``json.dumps`` never raises a ``TypeError``.
Args:
obj: Object to make JSON safe.
Returns:
A JSON-serializable version of the object.
"""
if obj is None or isinstance(obj, (str, int, float, bool)):
return obj
if isinstance(obj, (datetime, date)):
return obj.isoformat()
if is_dataclass(obj) and not isinstance(obj, type):
return make_json_safe(asdict(obj)) # type: ignore[arg-type]
if callable(getattr(obj, "model_dump", None)):
try:
return make_json_safe(obj.model_dump()) # type: ignore[no-any-return]
except TypeError:
pass
if callable(getattr(obj, "to_dict", None)):
try:
return make_json_safe(obj.to_dict()) # type: ignore[no-any-return]
except TypeError:
pass
if callable(getattr(obj, "dict", None)):
try:
return make_json_safe(obj.dict()) # type: ignore[no-any-return]
except TypeError:
pass
if isinstance(obj, dict):
return {str(key): make_json_safe(value) for key, value in obj.items()} # type: ignore[misc]
if isinstance(obj, (list, tuple)):
return [make_json_safe(item) for item in obj] # type: ignore[misc]
if hasattr(obj, "__dict__"):
return {key: make_json_safe(value) for key, value in vars(obj).items()} # type: ignore[misc]
return str(obj)
@@ -44,6 +44,7 @@ Only use skills from trusted sources.
from __future__ import annotations
import asyncio
import base64
import inspect
import json
import logging
@@ -60,6 +61,10 @@ from ._sessions import ContextProvider
from ._tools import FunctionTool
if TYPE_CHECKING:
from mcp.client.session import ClientSession
from mcp.types import ReadResourceResult
from pydantic import AnyUrl
from ._agents import SupportsAgentRun
from ._sessions import AgentSession, SessionContext
@@ -3285,4 +3290,443 @@ class AggregatingSkillsSource(SkillsSource):
return result
# region MCP Skills
def _mcp_any_url(uri: str) -> AnyUrl:
"""Convert a string URI to a :class:`pydantic.AnyUrl` for MCP client calls."""
from pydantic import AnyUrl as _AnyUrl
return _AnyUrl(uri)
def _is_mcp_resource_not_found(ex: Exception) -> bool:
"""Return ``True`` when *ex* is an :class:`McpError` indicating a missing resource.
Two codes are treated as "not found":
* ``-32002`` — the MCP-spec "Resource not found" code returned by a
compliant server when the URI does not exist. Not exported as a
constant from ``mcp.types`` but defined by the resources subprotocol.
* ``METHOD_NOT_FOUND`` (``-32601``) — the server does not implement
``resources/read`` at all, which for the skills source is functionally
equivalent to "no skills available."
All other codes — ``INVALID_PARAMS``, ``INTERNAL_ERROR``, ``PARSE_ERROR``,
``CONNECTION_CLOSED``, auth rejections, and generic handler errors
(code ``0``) — are treated as real failures so that a misconfigured
token or crashing server is not silently mistaken for "the server has no
skills."
"""
from mcp.shared.exceptions import McpError as _McpError
if not isinstance(ex, _McpError):
return False
from mcp.types import METHOD_NOT_FOUND as _METHOD_NOT_FOUND
return ex.error.code in {-32002, _METHOD_NOT_FOUND}
def _mcp_join_text(result: ReadResourceResult) -> str:
"""Join all :class:`TextResourceContents` items in a result into a single string."""
from mcp.types import TextResourceContents as _TextResourceContents
return "\n".join(c.text for c in result.contents if isinstance(c, _TextResourceContents))
class _McpSkillIndexEntry: # noqa: B903
"""A single entry in the ``skill://index.json`` discovery document.
All fields are optional to support lenient deserialization; callers
validate required fields before use.
"""
def __init__(
self,
*,
name: str | None = None,
type: str | None = None,
description: str | None = None,
url: str | None = None,
digest: str | None = None,
) -> None:
self.name = name
self.type = type
self.description = description
self.url = url
self.digest = digest
class _McpSkillIndex:
"""DTO for the ``skill://index.json`` discovery document.
Represents the Agent Skills Discovery v0.2.0 schema as bound to MCP
by SEP-2640.
"""
def __init__(
self,
*,
schema: str | None = None,
skills: list[_McpSkillIndexEntry] | None = None,
) -> None:
self.schema = schema
self.skills: list[_McpSkillIndexEntry] = skills if skills is not None else []
def _parse_mcp_skill_index(text: str) -> _McpSkillIndex:
"""Parse a JSON string into a :class:`_McpSkillIndex`.
Args:
text: Raw JSON text from ``skill://index.json``.
Returns:
A populated :class:`_McpSkillIndex` instance.
Raises:
json.JSONDecodeError: If the text is not valid JSON.
ValueError: If the top-level value is not a JSON object.
"""
raw: dict[str, Any] = json.loads(text)
if not isinstance(raw, dict):
raise ValueError("skill://index.json must be a JSON object")
entries: list[_McpSkillIndexEntry] = []
raw_skills: list[Any] = raw.get("skills") or []
for item in raw_skills:
if isinstance(item, dict):
d = cast(dict[str, Any], item)
entries.append(
_McpSkillIndexEntry(
name=d.get("name"),
type=d.get("type"),
description=d.get("description"),
url=d.get("url"),
digest=d.get("digest"),
)
)
return _McpSkillIndex(schema=raw.get("$schema"), skills=entries)
@experimental(feature_id=ExperimentalFeature.MCP_SKILLS)
class MCPSkillResource(SkillResource):
"""A :class:`SkillResource` backed by content fetched from an MCP server.
The :class:`~mcp.types.ReadResourceResult` is fetched eagerly by
:meth:`MCPSkill.get_resource` at construction time; :meth:`read`
extracts text or binary content from the result.
"""
def __init__(self, *, name: str, result: ReadResourceResult) -> None:
"""Initialize an MCPSkillResource.
Args:
name: The resource name (e.g. a relative path or identifier).
result: The result returned by the MCP server's ``resources/read`` request.
"""
super().__init__(name=name)
self._result = result
async def read(self, **kwargs: Any) -> Any:
"""Read the resource content.
Returns:
A ``bytes`` object when the resource contains binary content,
a ``str`` when it contains text, or ``None`` when the server
returned no content blocks.
"""
from mcp.types import BlobResourceContents, TextResourceContents
for content in self._result.contents:
if isinstance(content, BlobResourceContents):
blob = content.blob
# Strip data-URI prefix if present (some MCP servers send
# full data URIs instead of raw base64).
if blob.startswith("data:"):
blob = blob.split(",", 1)[-1]
return base64.b64decode(blob)
text = "\n".join(c.text for c in self._result.contents if isinstance(c, TextResourceContents))
return text if text else None
@experimental(feature_id=ExperimentalFeature.MCP_SKILLS)
class MCPSkill(Skill):
"""A :class:`Skill` discovered from an MCP server exposing the Agent Skills convention.
The skill is constructed from ``skill://index.json`` discovery metadata;
:meth:`get_content` fetches the full ``SKILL.md`` content from the MCP
server on demand via ``resources/read``.
Per SEP-2640, resources referenced inside SKILL.md are fetched on demand
via the originating MCP server: :meth:`get_resource` resolves a relative
resource name against the skill's root URI, issues a ``resources/read``
request, and returns an :class:`MCPSkillResource` with pre-fetched content.
"""
_SKILL_MD_SUFFIX: Final[str] = "SKILL.md"
def __init__(
self,
frontmatter: SkillFrontmatter,
skill_md_uri: str,
client: ClientSession,
) -> None:
"""Initialize an MCPSkill.
Args:
frontmatter: The parsed frontmatter metadata for this skill.
skill_md_uri: The full MCP resource URI of the ``SKILL.md`` resource
(e.g. ``skill://unit-converter/SKILL.md``). The skill's root URI
is derived by stripping the trailing ``SKILL.md`` segment.
client: The MCP client session used to fetch resources on demand.
"""
self._frontmatter = frontmatter
self._skill_md_uri = skill_md_uri
self._skill_root_uri = self._compute_skill_root_uri(skill_md_uri)
self._client = client
self._content: str | None = None
@property
def frontmatter(self) -> SkillFrontmatter:
"""The L1 discovery metadata for this skill."""
return self._frontmatter
async def get_content(self) -> str:
"""Get the full SKILL.md content from the MCP server.
Fetches the content via ``resources/read`` on the first call and
caches the result for subsequent calls.
Returns:
The SKILL.md content string.
Raises:
ValueError: If the MCP server returned no text content for the
SKILL.md resource.
"""
if self._content is not None:
return self._content
result = await self._client.read_resource(_mcp_any_url(self._skill_md_uri))
text = _mcp_join_text(result)
if not text:
raise ValueError(
f"The MCP server returned no text content for SKILL.md resource '{self._skill_md_uri}'."
)
self._content = text
return text
async def get_resource(self, name: str) -> SkillResource | None:
"""Get a sibling resource by name from the MCP server.
Resolves *name* as a relative path against the skill's root URI,
issues a ``resources/read`` request to the MCP server, and returns
an :class:`MCPSkillResource` with the pre-fetched content.
Args:
name: The resource name (e.g. ``references/checklist.md``).
Returns:
An :class:`MCPSkillResource`, or ``None`` when the name is empty
or the resource does not exist on the server.
"""
if not name or not name.strip():
return None
normalized = self._validate_resource_name(name)
if normalized is None:
return None
uri = self._skill_root_uri + normalized
try:
result = await self._client.read_resource(_mcp_any_url(uri))
except Exception as ex:
if _is_mcp_resource_not_found(ex):
logger.debug("MCP resource '%s' not available: %s", uri, ex)
return None
raise
return MCPSkillResource(name=name, result=result)
@staticmethod
def _validate_resource_name(name: str) -> str | None:
"""Validate a resource name and return the normalized form.
Defense in depth: refuses names that could escape the skill root
(absolute paths, embedded URI schemes, parent-traversal segments).
The MCP server is the authority on URI resolution, but rejecting
obviously unsafe shapes client-side avoids leaking escape attempts
upstream.
Args:
name: The raw resource name to validate.
Returns:
The normalized name with backslashes replaced by forward slashes,
or ``None`` if the name is unsafe.
"""
normalized = name.replace("\\", "/")
if (
normalized.startswith("/")
or "://" in normalized
or any(seg == ".." for seg in normalized.split("/"))
):
logger.debug("Rejecting resource name with unsafe path components: %r", name)
return None
return normalized
@staticmethod
def _compute_skill_root_uri(skill_md_uri: str) -> str:
"""Strip the trailing ``SKILL.md`` from the URI to produce the skill root.
If the URI doesn't end with ``SKILL.md``, ensures it ends with a
trailing slash.
"""
if skill_md_uri.endswith(MCPSkill._SKILL_MD_SUFFIX):
return skill_md_uri[: -len(MCPSkill._SKILL_MD_SUFFIX)]
if skill_md_uri.endswith("/"):
return skill_md_uri
return skill_md_uri + "/"
@experimental(feature_id=ExperimentalFeature.MCP_SKILLS)
class MCPSkillsSource(SkillsSource):
"""A :class:`SkillsSource` that discovers Agent Skills served over MCP.
Discovery follows the SEP-2640 recommended approach: the source reads
the well-known ``skill://index.json`` resource and constructs one
:class:`MCPSkill` per ``skill-md`` entry directly from the entry's
``name``, ``description``, and ``url`` fields.
The referenced ``SKILL.md`` resource is **not** read during discovery;
the host fetches its body on demand via ``resources/read`` when the
skill content is needed.
Only index entries of type ``skill-md`` are supported; entries of any
other type are silently skipped.
If ``skill://index.json`` is absent, unreadable, empty, or fails to
parse, this source returns an empty list.
Examples:
.. code-block:: python
from mcp.client.session import ClientSession
source = MCPSkillsSource(client=session)
skills = await source.get_skills()
"""
_INDEX_URI: Final[str] = "skill://index.json"
_SKILL_MD_TYPE: Final[str] = "skill-md"
def __init__(self, client: ClientSession) -> None:
"""Initialize an MCPSkillsSource.
Args:
client: An MCP client session connected to a server that
exposes Agent Skills resources.
"""
self._client = client
async def get_skills(self) -> list[Skill]:
"""Discover and return skills from the MCP server.
Reads ``skill://index.json``, parses it, and creates an
:class:`MCPSkill` for each valid ``skill-md`` entry.
Returns:
A list of discovered :class:`MCPSkill` instances.
"""
index = await self._try_read_index()
if index is None:
return []
skills: list[Skill] = []
for entry in index.skills:
result = self._try_create_skill(entry)
if result is not None:
skills.append(result)
logger.info("Loaded MCP skill: %s", result.frontmatter.name)
else:
logger.debug(
"Skipping skill index entry '%s'",
entry.name or "(unnamed)",
)
logger.info("Successfully loaded %d skills from MCP server", len(skills))
return skills
async def _try_read_index(self) -> _McpSkillIndex | None:
"""Attempt to read and parse ``skill://index.json`` from the MCP server.
Returns:
A parsed :class:`_McpSkillIndex`, or ``None`` if the index is
absent, empty, or malformed.
"""
try:
result = await self._client.read_resource(_mcp_any_url(self._INDEX_URI))
except Exception as ex:
if _is_mcp_resource_not_found(ex):
logger.debug("No skill://index.json resource available on MCP server: %s", ex)
return None
logger.warning("Failed to read skill://index.json from MCP server.", exc_info=True)
raise
index_text = _mcp_join_text(result)
if not index_text:
logger.debug("skill://index.json on MCP server returned empty/non-text contents")
return None
try:
return _parse_mcp_skill_index(index_text)
except (json.JSONDecodeError, ValueError):
logger.warning("Failed to parse skill://index.json JSON document.", exc_info=True)
return None
def _try_create_skill(self, entry: _McpSkillIndexEntry) -> MCPSkill | None:
"""Attempt to create an :class:`MCPSkill` from an index entry.
Args:
entry: A single entry from the skill index.
Returns:
An :class:`MCPSkill` if the entry is valid, or ``None`` if the
entry should be skipped.
"""
if entry.type != self._SKILL_MD_TYPE:
logger.debug(
"Skipping entry '%s': unsupported type '%s'",
entry.name or "(unnamed)",
entry.type or "(none)",
)
return None
if not entry.name or not entry.name.strip():
logger.debug("Skipping entry: missing required 'name' field")
return None
if not entry.description or not entry.description.strip():
logger.debug("Skipping entry '%s': missing required 'description' field", entry.name)
return None
if not entry.url or not entry.url.strip():
logger.debug("Skipping entry '%s': missing required 'url' field", entry.name)
return None
try:
fm = SkillFrontmatter(name=entry.name, description=entry.description)
except ValueError as ex:
logger.debug("Skipping entry '%s': invalid metadata: %s", entry.name, ex)
return None
return MCPSkill(frontmatter=fm, skill_md_uri=entry.url, client=self._client)
# endregion
+38 -4
View File
@@ -292,6 +292,7 @@ class FunctionTool(SerializationMixin):
"_cached_parameters",
"_input_schema",
"_schema_supplied",
"_invoke_sync_on_event_loop",
}
def __init__(
@@ -366,6 +367,7 @@ class FunctionTool(SerializationMixin):
self.description = description
self.kind = kind
self.additional_properties = additional_properties
self._invoke_sync_on_event_loop = False
for key, value in kwargs.items():
setattr(self, key, value)
@@ -537,6 +539,16 @@ class FunctionTool(SerializationMixin):
self.invocation_exception_count += 1
raise
async def _invoke_function(self, call_kwargs: Mapping[str, Any]) -> Any:
"""Run sync tools off the event loop during async invocation."""
func = self.func.func if isinstance(self.func, FunctionTool) else self.func
if inspect.iscoroutinefunction(func) or getattr(self, "_invoke_sync_on_event_loop", False):
res = self.__call__(**call_kwargs)
return await res if inspect.isawaitable(res) else res
res = await asyncio.to_thread(self.__call__, **call_kwargs)
return await res if inspect.isawaitable(res) else res
@overload
async def invoke(
self,
@@ -679,8 +691,7 @@ class FunctionTool(SerializationMixin):
if not OBSERVABILITY_SETTINGS.ENABLED: # type: ignore[name-defined]
logger.info(f"Function name: {self.name}")
logger.debug(f"Function arguments: {observable_kwargs}")
res = self.__call__(**call_kwargs)
result = await res if inspect.isawaitable(res) else res
result = await self._invoke_function(call_kwargs)
if skip_parsing:
logger.info(f"Function {self.name} succeeded.")
logger.debug(f"Function result: {type(result).__name__}")
@@ -730,8 +741,7 @@ class FunctionTool(SerializationMixin):
start_time_stamp = perf_counter()
end_time_stamp: float | None = None
try:
res = self.__call__(**call_kwargs)
result = await res if inspect.isawaitable(res) else res
result = await self._invoke_function(call_kwargs)
end_time_stamp = perf_counter()
except Exception as exception:
end_time_stamp = perf_counter()
@@ -1418,6 +1428,7 @@ async def _auto_invoke_function(
sequence_index: int | None = None,
request_index: int | None = None,
middleware_pipeline: FunctionMiddlewarePipeline | None = None,
live_tools: list[ToolTypes] | None = None,
) -> Content:
"""Invoke a function call requested by the agent, applying middleware that is defined.
@@ -1432,6 +1443,8 @@ async def _auto_invoke_function(
sequence_index: The index of the function call in the sequence.
request_index: The index of the request iteration.
middleware_pipeline: Optional middleware pipeline to apply during execution.
live_tools: The live, mutable tools list for the current agent run, exposed on
the FunctionInvocationContext so tools can add/remove tools at runtime.
Returns:
The function result content.
@@ -1523,6 +1536,7 @@ async def _auto_invoke_function(
arguments=args,
session=invocation_session,
kwargs=runtime_kwargs.copy(),
tools=live_tools,
)
function_result = await tool.invoke(
arguments=args,
@@ -1537,6 +1551,10 @@ async def _auto_invoke_function(
except UserInputRequiredException:
raise
except Exception as exc:
logger.warning(
f"Function '{tool.name}' raised an exception; returning an error result to the "
f"model. Set include_detailed_errors=True for the full detail. Exception: {exc!r}"
)
message = "Error: Function failed."
if config.get("include_detailed_errors", False):
message = f"{message} Exception: {exc}"
@@ -1552,6 +1570,7 @@ async def _auto_invoke_function(
arguments=args,
session=invocation_session,
kwargs=runtime_kwargs.copy(),
tools=live_tools,
)
call_id = function_call_content.call_id
@@ -1608,6 +1627,10 @@ async def _auto_invoke_function(
except UserInputRequiredException:
raise
except Exception as exc:
logger.warning(
f"Function '{tool.name}' raised an exception; returning an error result to the "
f"model. Set include_detailed_errors=True for the full detail. Exception: {exc!r}"
)
message = "Error: Function failed."
if config.get("include_detailed_errors", False):
message = f"{message} Exception: {exc}"
@@ -1659,6 +1682,9 @@ async def _try_execute_function_calls(
from ._types import Content
tool_map = _get_tool_map(tools)
# The live tools list (when tools is the run-local list) is exposed on the
# FunctionInvocationContext so tools can add/remove tools during the run.
live_tools: list[ToolTypes] | None = cast("list[ToolTypes]", tools) if isinstance(tools, list) else None
approval_tools = [tool_name for tool_name, tool in tool_map.items() if tool.approval_mode == "always_require"]
logger.debug(
"_try_execute_function_calls: tool_map keys=%s, approval_tools=%s",
@@ -1733,6 +1759,7 @@ async def _try_execute_function_calls(
request_index=attempt_idx,
middleware_pipeline=middleware_pipeline,
config=config,
live_tools=live_tools,
)
return (result, False)
except MiddlewareTermination as exc:
@@ -2371,6 +2398,13 @@ class FunctionInvocationLayer(Generic[OptionsCoT]):
function_invocation_kwargs=function_invocation_kwargs,
client_kwargs=filtered_kwargs,
)
# Establish a single, run-local mutable tools list so that tools can add or remove
# tools during the run (progressive tool exposure). A fresh list is created via
# normalize_tools so the caller's original tools container is never mutated, while
# the same list object is shared with the model (options["tools"]) and the tool map
# rebuilt on every loop iteration.
if mutable_options.get("tools"):
mutable_options["tools"] = normalize_tools(mutable_options["tools"])
if not stream:
async def _get_response() -> ChatResponse[Any]:
@@ -12,6 +12,7 @@ from datetime import datetime, timezone
from typing import TYPE_CHECKING, Any, ClassVar, Literal, cast, overload
from .._agents import BaseAgent
from .._serialization import make_json_safe
from .._sessions import (
AgentSession,
ContextProvider,
@@ -61,7 +62,7 @@ class WorkflowAgent(BaseAgent):
data: Any
def to_dict(self) -> dict[str, Any]:
return {"request_id": self.request_id, "data": self.data}
return {"request_id": self.request_id, "data": make_json_safe(self.data)}
def to_json(self) -> str:
return json.dumps(self.to_dict())
@@ -47,6 +47,7 @@ from copy import deepcopy
from typing import Any, Generic, Literal, TypeVar, overload
from .._feature_stage import ExperimentalFeature, experimental
from .._serialization import make_json_safe
from .._types import AgentResponse, AgentResponseUpdate, ResponseStream
from ..observability import OtelAttr, capture_exception, create_workflow_span
from ._checkpoint import CheckpointStorage, WorkflowCheckpoint
@@ -1515,7 +1516,7 @@ class FunctionalWorkflowAgent:
function_call = Content.from_function_call(
call_id=request_id,
name=self.REQUEST_INFO_FUNCTION_NAME,
arguments={"request_id": request_id, "data": event.data},
arguments={"request_id": request_id, "data": make_json_safe(event.data)},
)
return Content.from_function_approval_request(
id=request_id,
@@ -34,6 +34,7 @@ _IMPORTS: dict[str, tuple[str, str]] = {
"FoundryLocalChatOptions": ("agent_framework_foundry_local", "agent-framework-foundry-local"),
"FoundryLocalClient": ("agent_framework_foundry_local", "agent-framework-foundry-local"),
"FoundryLocalSettings": ("agent_framework_foundry_local", "agent-framework-foundry-local"),
"GeneratedEvaluatorRef": ("agent_framework_foundry", "agent-framework-foundry"),
"RawAnthropicFoundryClient": ("agent_framework_anthropic", "agent-framework-anthropic"),
"RawFoundryAgent": ("agent_framework_foundry", "agent-framework-foundry"),
"RawFoundryAgentChatClient": ("agent_framework_foundry", "agent-framework-foundry"),
@@ -20,6 +20,7 @@ from agent_framework_foundry import (
FoundryEmbeddingSettings,
FoundryEvals,
FoundryMemoryProvider,
GeneratedEvaluatorRef,
RawFoundryAgent,
RawFoundryAgentChatClient,
RawFoundryChatClient,
@@ -52,6 +53,7 @@ __all__ = [
"FoundryLocalClient",
"FoundryLocalSettings",
"FoundryMemoryProvider",
"GeneratedEvaluatorRef",
"RawAnthropicFoundryClient",
"RawFoundryAgent",
"RawFoundryAgentChatClient",
@@ -498,14 +498,34 @@ def _get_exporters_from_env(
# Get base endpoint
base_endpoint = os.getenv("OTEL_EXPORTER_OTLP_ENDPOINT")
# Get signal-specific endpoints (these override base endpoint)
traces_endpoint = os.getenv("OTEL_EXPORTER_OTLP_TRACES_ENDPOINT") or base_endpoint
metrics_endpoint = os.getenv("OTEL_EXPORTER_OTLP_METRICS_ENDPOINT") or base_endpoint
logs_endpoint = os.getenv("OTEL_EXPORTER_OTLP_LOGS_ENDPOINT") or base_endpoint
# Get signal-specific endpoints (these override base endpoint and are used verbatim)
traces_endpoint_specific = os.getenv("OTEL_EXPORTER_OTLP_TRACES_ENDPOINT")
metrics_endpoint_specific = os.getenv("OTEL_EXPORTER_OTLP_METRICS_ENDPOINT")
logs_endpoint_specific = os.getenv("OTEL_EXPORTER_OTLP_LOGS_ENDPOINT")
# Get protocol (default is grpc)
protocol = os.getenv("OTEL_EXPORTER_OTLP_PROTOCOL", "grpc").lower()
# Per the OTel spec, OTEL_EXPORTER_OTLP_ENDPOINT is a *base* URL for HTTP — the SDK
# auto-appends /v1/{traces,metrics,logs} when it reads the env var directly. The
# signal-specific endpoint env vars are *full* URLs used verbatim. Because we read
# the env vars here and forward them as the ``endpoint=`` constructor argument
# (which the SDK always treats as a full URL), we must replicate the auto-append
# ourselves for HTTP when falling back to the base endpoint. For gRPC, the base
# endpoint is used as-is.
traces_endpoint: str | None
metrics_endpoint: str | None
logs_endpoint: str | None
if protocol in ("http/protobuf", "http") and base_endpoint:
base_for_http = base_endpoint.rstrip("/")
traces_endpoint = traces_endpoint_specific or f"{base_for_http}/v1/traces"
metrics_endpoint = metrics_endpoint_specific or f"{base_for_http}/v1/metrics"
logs_endpoint = logs_endpoint_specific or f"{base_for_http}/v1/logs"
else:
traces_endpoint = traces_endpoint_specific or base_endpoint
metrics_endpoint = metrics_endpoint_specific or base_endpoint
logs_endpoint = logs_endpoint_specific or base_endpoint
# Get base headers
base_headers_str = os.getenv("OTEL_EXPORTER_OTLP_HEADERS", "")
base_headers = _parse_headers(base_headers_str)
+1 -1
View File
@@ -4,7 +4,7 @@ description = "Microsoft Agent Framework for building AI Agents with Python. Thi
authors = [{ name = "Microsoft", email = "af-support@microsoft.com"}]
readme = "README.md"
requires-python = ">=3.10"
version = "1.7.0"
version = "1.8.0"
license-files = ["LICENSE"]
urls.homepage = "https://aka.ms/agent-framework"
urls.source = "https://github.com/microsoft/agent-framework/tree/main/python"
@@ -11,10 +11,14 @@ from agent_framework import (
GROUP_TOKEN_COUNT_KEY,
BaseChatClient,
ChatResponse,
ChatResponseUpdate,
Content,
Message,
SlidingWindowStrategy,
SupportsChatGetResponse,
ToolResultCompactionStrategy,
TruncationStrategy,
tool,
)
@@ -258,6 +262,196 @@ async def test_base_client_default_tokenizer_without_strategy_annotates_messages
assert captured_token_counts == [[19, 19]]
def _tool_call_response(call_id: str, location: str) -> ChatResponse:
return ChatResponse(
messages=Message(
role="assistant",
contents=[
Content.from_function_call(
call_id=call_id,
name="lookup_weather",
arguments=f'{{"location": "{location}"}}',
)
],
),
response_id=f"resp_{call_id}",
)
def _is_tool_result_summary(message: Message) -> bool:
text = message.text or ""
return message.role == "assistant" and text.startswith("[Tool results:")
async def test_function_loop_persists_inserted_summaries_across_iterations(
chat_client_base: SupportsChatGetResponse,
) -> None:
# Regression test for #4991: compaction inserts summary messages and excludes the
# originals. Across tool-loop iterations the exclusion flags persisted (shared Message
# objects) but the inserted summaries were dropped (they only lived on a throwaway copy),
# so older tool groups were silently lost with no summary representing them.
chat_client_base.function_invocation_configuration["enabled"] = True # type: ignore[attr-defined]
chat_client_base.function_invocation_configuration["max_iterations"] = 3 # type: ignore[attr-defined]
chat_client_base.compaction_strategy = ToolResultCompactionStrategy(keep_last_tool_call_groups=1) # type: ignore[attr-defined]
@tool(name="lookup_weather", approval_mode="never_require")
def lookup_weather(location: str) -> str:
return f"Weather in {location}: sunny"
chat_client_base.run_responses = [ # type: ignore[attr-defined]
_tool_call_response("call_1", "London"),
_tool_call_response("call_2", "Paris"),
_tool_call_response("call_3", "Tokyo"),
]
captured_inputs: list[list[Message]] = []
original = chat_client_base._get_non_streaming_response # type: ignore[attr-defined]
async def _capture(
*,
messages: list[Message],
options: dict[str, Any],
**kwargs: Any,
) -> ChatResponse:
captured_inputs.append(list(messages))
return await original(messages=messages, options=options, **kwargs)
chat_client_base._get_non_streaming_response = _capture # type: ignore[attr-defined,method-assign]
await chat_client_base.get_response(
[Message(role="user", contents=["What is the weather in London?"])],
options={"tools": [lookup_weather]}, # type: ignore[typeddict-unknown-key]
)
# The final model call should represent every compacted tool group with a summary.
# Two older tool groups get collapsed (London, Paris) while the last (Tokyo) is kept.
final_input = captured_inputs[-1]
summaries = [message for message in final_input if _is_tool_result_summary(message)]
summary_text = " ".join(message.text or "" for message in summaries)
assert len(summaries) == 2, [message.text for message in final_input]
assert "London" in summary_text
assert "Paris" in summary_text
def _tool_call_update(call_id: str, location: str) -> list[ChatResponseUpdate]:
return [
ChatResponseUpdate(
contents=[
Content.from_function_call(
call_id=call_id,
name="lookup_weather",
arguments=f'{{"location": "{location}"}}',
)
],
role="assistant",
finish_reason="stop",
response_id=f"resp_{call_id}",
)
]
async def test_function_loop_persists_inserted_summaries_across_iterations_streaming(
chat_client_base: SupportsChatGetResponse,
) -> None:
# Streaming counterpart of the #4991 regression test: the summary persistence fix in
# ``_prepare_messages_for_model_call`` must cover the streaming tool loop too.
chat_client_base.function_invocation_configuration["enabled"] = True # type: ignore[attr-defined]
chat_client_base.function_invocation_configuration["max_iterations"] = 3 # type: ignore[attr-defined]
chat_client_base.compaction_strategy = ToolResultCompactionStrategy(keep_last_tool_call_groups=1) # type: ignore[attr-defined]
@tool(name="lookup_weather", approval_mode="never_require")
def lookup_weather(location: str) -> str:
return f"Weather in {location}: sunny"
chat_client_base.streaming_responses = [ # type: ignore[attr-defined]
_tool_call_update("call_1", "London"),
_tool_call_update("call_2", "Paris"),
_tool_call_update("call_3", "Tokyo"),
]
captured_inputs: list[list[Message]] = []
original = chat_client_base._get_streaming_response # type: ignore[attr-defined]
def _capture(
*,
messages: list[Message],
options: dict[str, Any],
**kwargs: Any,
):
captured_inputs.append(list(messages))
return original(messages=messages, options=options, **kwargs)
chat_client_base._get_streaming_response = _capture # type: ignore[attr-defined,method-assign]
stream = chat_client_base.get_response(
[Message(role="user", contents=["What is the weather in London?"])],
stream=True,
options={"tools": [lookup_weather]}, # type: ignore[typeddict-unknown-key]
)
async for _ in stream:
pass
final_input = captured_inputs[-1]
summaries = [message for message in final_input if _is_tool_result_summary(message)]
summary_text = " ".join(message.text or "" for message in summaries)
assert len(summaries) == 2, [message.text for message in final_input]
assert "London" in summary_text
assert "Paris" in summary_text
async def test_function_loop_compaction_conversation_id_mode_does_not_resend_history(
chat_client_base: SupportsChatGetResponse,
) -> None:
# In conversation-id mode the server owns prior context, so the tool loop clears
# ``prepped_messages`` and only sends the latest message. Compaction must not fight that
# by re-inserting summaries or re-sending earlier turns.
chat_client_base.function_invocation_configuration["enabled"] = True # type: ignore[attr-defined]
chat_client_base.function_invocation_configuration["max_iterations"] = 3 # type: ignore[attr-defined]
chat_client_base.compaction_strategy = ToolResultCompactionStrategy(keep_last_tool_call_groups=1) # type: ignore[attr-defined]
@tool(name="lookup_weather", approval_mode="never_require")
def lookup_weather(location: str) -> str:
return f"Weather in {location}: sunny"
def _conversation_tool_call(call_id: str, location: str) -> ChatResponse:
response = _tool_call_response(call_id, location)
response.conversation_id = "conv_1"
return response
chat_client_base.run_responses = [ # type: ignore[attr-defined]
_conversation_tool_call("call_1", "London"),
_conversation_tool_call("call_2", "Paris"),
_conversation_tool_call("call_3", "Tokyo"),
]
captured_inputs: list[list[Message]] = []
original = chat_client_base._get_non_streaming_response # type: ignore[attr-defined]
async def _capture(
*,
messages: list[Message],
options: dict[str, Any],
**kwargs: Any,
) -> ChatResponse:
captured_inputs.append(list(messages))
return await original(messages=messages, options=options, **kwargs)
chat_client_base._get_non_streaming_response = _capture # type: ignore[attr-defined,method-assign]
await chat_client_base.get_response(
[Message(role="user", contents=["What is the weather in London?"])],
options={"tools": [lookup_weather]}, # type: ignore[typeddict-unknown-key]
)
# After the conversation id is established the loop only forwards the latest message,
# so subsequent model calls never receive the full history or summary messages.
for sent in captured_inputs[1:]:
assert len(sent) <= 1, [message.text for message in sent]
assert not any(_is_tool_result_summary(message) for message in sent)
def test_base_client_as_agent_does_not_copy_client_compaction_defaults(
chat_client_base: SupportsChatGetResponse,
) -> None:
@@ -196,6 +196,64 @@ def test_append_compaction_message_annotates_new_message() -> None:
assert isinstance(_group_id(messages[1]), str)
def test_incremental_annotation_assigns_unique_message_ids() -> None:
# Regression test for #5237: ``_ensure_message_ids`` assigned ``msg_{index}``
# using the position within the slice handed to ``group_messages``. Successive
# incremental annotations restart the index at 0, so distinct messages collided
# on the same ``message_id``.
messages: list[Message] = []
for turn in range(4):
messages.append(Message(role="user", contents=[f"user {turn}"]))
annotate_message_groups(messages)
messages.append(Message(role="assistant", contents=[f"assistant {turn}"]))
annotate_message_groups(messages)
message_ids = [message.message_id for message in messages]
assert all(message_ids), "every message should receive an id"
assert len(set(message_ids)) == len(message_ids), f"duplicate message ids: {message_ids}"
def test_ensure_message_ids_avoids_existing_id_collisions() -> None:
# An auto-generated ``msg_{index}`` must not collide with an id already present
# on another message (user-supplied or assigned by an earlier annotation pass).
messages = [
Message(role="user", contents=["zero"]),
Message(role="assistant", contents=["one"], message_id="msg_2"),
Message(role="user", contents=["two"]),
]
annotate_message_groups(messages)
message_ids = [message.message_id for message in messages]
assert message_ids[1] == "msg_2"
assert len(set(message_ids)) == len(message_ids), f"duplicate message ids: {message_ids}"
def test_incremental_annotation_avoids_prefix_id_collision() -> None:
# Regression for the PR review on #5237: when only a suffix is re-annotated,
# an auto-assigned ``msg_{index}`` in the suffix must not collide with a
# preexisting id carried by a message in the *preserved prefix* (a group
# before the one re-annotation pulls back to). Otherwise ``_group_id_for``
# derives the same group id and merges groups across the boundary.
messages = [
# Out-of-position, user-supplied id that matches the ``msg_{index}`` the
# suffix pass would assign to the appended message below. This message is
# two groups back, so it stays outside the re-annotated slice.
Message(role="user", contents=["zero"], message_id="msg_2"),
Message(role="user", contents=["one"]),
]
annotate_message_groups(messages)
assert messages[0].message_id == "msg_2"
assert messages[1].message_id == "msg_1"
messages.append(Message(role="user", contents=["two"]))
annotate_message_groups(messages, from_index=2)
message_ids = [message.message_id for message in messages]
assert all(message_ids), "every message should receive an id"
assert len(set(message_ids)) == len(message_ids), f"duplicate message ids: {message_ids}"
assert messages[0].message_id == "msg_2"
async def test_truncation_strategy_keeps_system_anchor() -> None:
messages = [
Message(role="system", contents=["you are helpful"]),
@@ -484,6 +542,44 @@ async def test_tool_result_compaction_collapses_old_groups_into_summary() -> Non
assert any(m.role == "tool" for m in projected)
async def test_tool_result_compaction_is_idempotent_after_summary_insertion() -> None:
"""Re-running compaction after a mid-list summary insertion must not duplicate it.
Mirrors a subsequent tool-loop iteration (issue #4991): the inserted summary and the
excluded originals now persist on the same list, so a second annotate + compaction pass
over the same groups should be a no-op rather than collapsing the group again.
"""
messages = [
Message(role="user", contents=["u"]),
_assistant_function_call("call-1"),
_tool_result("call-1", "r1"),
_assistant_function_call("call-2"),
_tool_result("call-2", "r2"),
Message(role="assistant", contents=["done"]),
]
strategy = ToolResultCompactionStrategy(keep_last_tool_call_groups=1)
annotate_message_groups(messages)
assert await strategy(messages) is True
summaries_after_first = [m for m in messages if (m.text or "").startswith("[Tool results:")]
assert len(summaries_after_first) == 1
summary = summaries_after_first[0]
summary_group_ids = _group_unknown_value(summary, SUMMARY_OF_GROUP_IDS_KEY)
# Second pass over the same (now partially compacted) list.
annotate_message_groups(messages)
changed = await strategy(messages)
assert changed is False
summaries_after_second = [m for m in messages if (m.text or "").startswith("[Tool results:")]
assert len(summaries_after_second) == 1
assert _group_unknown_value(summaries_after_second[0], SUMMARY_OF_GROUP_IDS_KEY) == summary_group_ids
# The kept tool-call group stays atomic and included.
projected = included_messages(messages)
assert any(m.role == "tool" for m in projected)
async def test_tool_result_compaction_zero_collapses_all() -> None:
"""With keep=0, all tool-call groups are collapsed into summaries."""
messages = [
@@ -3975,3 +3975,425 @@ async def test_user_input_request_empty_contents_returns_fallback(chat_client_ba
]
assert len(function_results) >= 1
assert any("user input" in (fr.result or "").lower() for fr in function_results)
# region Progressive tool exposure (FunctionInvocationContext.add_tools / remove_tools)
def _pte_function_call_response(call_id: str, name: str, arguments: str = "{}") -> ChatResponse:
return ChatResponse(
messages=Message(
role="assistant",
contents=[Content.from_function_call(call_id=call_id, name=name, arguments=arguments)],
)
)
def _pte_text_response(text: str = "done") -> ChatResponse:
return ChatResponse(messages=Message(role="assistant", contents=[text]))
@tool(name="factorial", approval_mode="never_require")
def _pte_factorial(n: int) -> int:
"""Compute the factorial of n."""
result = 1
for value in range(2, n + 1):
result *= value
return result
async def test_context_exposes_live_tools(chat_client_base: SupportsChatGetResponse):
from agent_framework import FunctionTool
seen_names: list[str] = []
@tool(name="inspect_tools", approval_mode="never_require")
def inspect_tools(ctx: FunctionInvocationContext) -> str:
assert ctx.tools is not None
seen_names.extend(t.name for t in ctx.tools if isinstance(t, FunctionTool))
return "inspected"
chat_client_base.run_responses = [
_pte_function_call_response("1", "inspect_tools"),
_pte_text_response(),
]
await chat_client_base.get_response(
[Message(role="user", contents=["hi"])],
options={"tool_choice": "auto", "tools": [inspect_tools]},
)
assert "inspect_tools" in seen_names
async def test_add_tools_available_next_iteration(chat_client_base: SupportsChatGetResponse):
exec_counter = 0
@tool(name="factorial", approval_mode="never_require")
def factorial(n: int) -> int:
nonlocal exec_counter
exec_counter += 1
return 120
@tool(name="load_math", approval_mode="never_require")
def load_math(ctx: FunctionInvocationContext) -> str:
ctx.add_tools(factorial)
return "math tools loaded"
chat_client_base.function_invocation_configuration["max_iterations"] = 3 # type: ignore[attr-defined]
chat_client_base.run_responses = [
_pte_function_call_response("1", "load_math"),
_pte_function_call_response("2", "factorial", '{"n": 5}'),
_pte_text_response(),
]
response = await chat_client_base.get_response(
[Message(role="user", contents=["compute 5!"])],
options={"tool_choice": "auto", "tools": [load_math]},
)
assert exec_counter == 1
assert response.messages[-1].text == "done"
async def test_add_tools_model_sees_added_tools_in_options(chat_client_base: SupportsChatGetResponse):
from agent_framework import FunctionTool
recorded: list[list[str]] = []
client_cls = type(chat_client_base)
original = client_cls._get_non_streaming_response
async def recording(self: Any, *, messages: Any, options: dict[str, Any], **kwargs: Any) -> ChatResponse:
tools = options.get("tools") or []
recorded.append([t.name for t in tools if isinstance(t, FunctionTool)])
return await original(self, messages=messages, options=options, **kwargs)
@tool(name="load_math", approval_mode="never_require")
def load_math(ctx: FunctionInvocationContext) -> str:
ctx.add_tools(_pte_factorial)
return "loaded"
chat_client_base.function_invocation_configuration["max_iterations"] = 3 # type: ignore[attr-defined]
chat_client_base.run_responses = [
_pte_function_call_response("1", "load_math"),
_pte_function_call_response("2", "factorial", '{"n": 5}'),
_pte_text_response(),
]
monkey = pytest.MonkeyPatch()
monkey.setattr(client_cls, "_get_non_streaming_response", recording)
try:
await chat_client_base.get_response(
[Message(role="user", contents=["compute 5!"])],
options={"tool_choice": "auto", "tools": [load_math]},
)
finally:
monkey.undo()
assert recorded[0] == ["load_math"]
assert "factorial" in recorded[1]
async def test_remove_tools_next_iteration(chat_client_base: SupportsChatGetResponse):
from agent_framework import FunctionTool
recorded: list[list[str]] = []
client_cls = type(chat_client_base)
original = client_cls._get_non_streaming_response
async def recording(self: Any, *, messages: Any, options: dict[str, Any], **kwargs: Any) -> ChatResponse:
tools = options.get("tools") or []
recorded.append([t.name for t in tools if isinstance(t, FunctionTool)])
return await original(self, messages=messages, options=options, **kwargs)
@tool(name="get_weather", approval_mode="never_require")
def get_weather(location: str) -> str:
return "sunny"
@tool(name="drop_weather", approval_mode="never_require")
def drop_weather(ctx: FunctionInvocationContext) -> str:
ctx.remove_tools("get_weather")
return "removed"
chat_client_base.function_invocation_configuration["max_iterations"] = 3 # type: ignore[attr-defined]
chat_client_base.run_responses = [
_pte_function_call_response("1", "drop_weather"),
_pte_text_response(),
]
monkey = pytest.MonkeyPatch()
monkey.setattr(client_cls, "_get_non_streaming_response", recording)
try:
await chat_client_base.get_response(
[Message(role="user", contents=["hi"])],
options={"tool_choice": "auto", "tools": [get_weather, drop_weather]},
)
finally:
monkey.undo()
assert set(recorded[0]) == {"get_weather", "drop_weather"}
assert "get_weather" not in recorded[1]
async def test_add_tools_does_not_mutate_caller_tools_list(chat_client_base: SupportsChatGetResponse):
@tool(name="load_math", approval_mode="never_require")
def load_math(ctx: FunctionInvocationContext) -> str:
ctx.add_tools(_pte_factorial)
return "loaded"
original_tools: list[Any] = [load_math]
chat_client_base.function_invocation_configuration["max_iterations"] = 3 # type: ignore[attr-defined]
chat_client_base.run_responses = [
_pte_function_call_response("1", "load_math"),
_pte_text_response(),
]
await chat_client_base.get_response(
[Message(role="user", contents=["hi"])],
options={"tool_choice": "auto", "tools": original_tools},
)
assert original_tools == [load_math]
async def test_add_tools_persists_across_iterations(chat_client_base: SupportsChatGetResponse):
from agent_framework import FunctionTool
recorded: list[list[str]] = []
client_cls = type(chat_client_base)
original = client_cls._get_non_streaming_response
async def recording(self: Any, *, messages: Any, options: dict[str, Any], **kwargs: Any) -> ChatResponse:
tools = options.get("tools") or []
recorded.append([t.name for t in tools if isinstance(t, FunctionTool)])
return await original(self, messages=messages, options=options, **kwargs)
@tool(name="load_math", approval_mode="never_require")
def load_math(ctx: FunctionInvocationContext) -> str:
ctx.add_tools(_pte_factorial)
return "loaded"
chat_client_base.function_invocation_configuration["max_iterations"] = 4 # type: ignore[attr-defined]
chat_client_base.run_responses = [
_pte_function_call_response("1", "load_math"),
_pte_function_call_response("2", "factorial", '{"n": 5}'),
_pte_function_call_response("3", "factorial", '{"n": 3}'),
_pte_text_response(),
]
monkey = pytest.MonkeyPatch()
monkey.setattr(client_cls, "_get_non_streaming_response", recording)
try:
await chat_client_base.get_response(
[Message(role="user", contents=["hi"])],
options={"tool_choice": "auto", "tools": [load_math]},
)
finally:
monkey.undo()
assert "factorial" in recorded[1]
assert "factorial" in recorded[2]
async def test_add_tools_through_function_middleware(chat_client_base: SupportsChatGetResponse):
exec_counter = 0
class PassthroughMiddleware(FunctionMiddleware):
async def process(self, context: FunctionInvocationContext, call_next: Any) -> None:
await call_next()
@tool(name="factorial", approval_mode="never_require")
def factorial(n: int) -> int:
nonlocal exec_counter
exec_counter += 1
return 120
@tool(name="load_math", approval_mode="never_require")
def load_math(ctx: FunctionInvocationContext) -> str:
ctx.add_tools(factorial)
return "loaded"
chat_client_base.function_invocation_configuration["max_iterations"] = 3 # type: ignore[attr-defined]
chat_client_base.run_responses = [
_pte_function_call_response("1", "load_math"),
_pte_function_call_response("2", "factorial", '{"n": 5}'),
_pte_text_response(),
]
await chat_client_base.get_response(
[Message(role="user", contents=["hi"])],
options={"tool_choice": "auto", "tools": [load_math]},
middleware=[PassthroughMiddleware()],
)
assert exec_counter == 1
async def test_add_tools_with_approval_required_tool(chat_client_base: SupportsChatGetResponse):
@tool(name="secure_tool", approval_mode="always_require")
def secure_tool(value: str) -> str:
return f"secure: {value}"
@tool(name="load_secure", approval_mode="never_require")
def load_secure(ctx: FunctionInvocationContext) -> str:
ctx.add_tools(secure_tool)
return "loaded"
chat_client_base.function_invocation_configuration["max_iterations"] = 3 # type: ignore[attr-defined]
chat_client_base.run_responses = [
_pte_function_call_response("1", "load_secure"),
_pte_function_call_response("2", "secure_tool", '{"value": "x"}'),
_pte_text_response(),
]
response = await chat_client_base.get_response(
[Message(role="user", contents=["hi"])],
options={"tool_choice": "auto", "tools": [load_secure]},
)
assert any(item.type == "function_approval_request" for msg in response.messages for item in msg.contents)
async def test_add_tools_accepts_plain_callable(chat_client_base: SupportsChatGetResponse):
exec_counter = 0
def plain_factorial(n: int) -> int:
"""Compute factorial."""
nonlocal exec_counter
exec_counter += 1
return 120
@tool(name="load_math", approval_mode="never_require")
def load_math(ctx: FunctionInvocationContext) -> str:
ctx.add_tools(plain_factorial)
return "loaded"
chat_client_base.function_invocation_configuration["max_iterations"] = 3 # type: ignore[attr-defined]
chat_client_base.run_responses = [
_pte_function_call_response("1", "load_math"),
_pte_function_call_response("2", "plain_factorial", '{"n": 5}'),
_pte_text_response(),
]
await chat_client_base.get_response(
[Message(role="user", contents=["hi"])],
options={"tool_choice": "auto", "tools": [load_math]},
)
assert exec_counter == 1
async def test_add_tools_streaming(chat_client_base: SupportsChatGetResponse):
exec_counter = 0
@tool(name="factorial", approval_mode="never_require")
def factorial(n: int) -> int:
nonlocal exec_counter
exec_counter += 1
return 120
@tool(name="load_math", approval_mode="never_require")
def load_math(ctx: FunctionInvocationContext) -> str:
ctx.add_tools(factorial)
return "loaded"
chat_client_base.function_invocation_configuration["max_iterations"] = 3 # type: ignore[attr-defined]
chat_client_base.streaming_responses = [
[
ChatResponseUpdate(
contents=[Content.from_function_call(call_id="1", name="load_math", arguments="{}")],
role="assistant",
)
],
[
ChatResponseUpdate(
contents=[Content.from_function_call(call_id="2", name="factorial", arguments='{"n": 5}')],
role="assistant",
)
],
[ChatResponseUpdate(contents=[Content.from_text("done")], role="assistant", finish_reason="stop")],
]
async for _ in chat_client_base.get_response(
[Message(role="user", contents=["hi"])],
stream=True,
options={"tool_choice": "auto", "tools": [load_math]},
):
pass
assert exec_counter == 1
def test_add_tools_duplicate_same_object_is_noop():
@tool(name="dup", approval_mode="never_require")
def dup(x: int) -> int:
return x
ctx = FunctionInvocationContext(function=dup, arguments={}, tools=[dup])
ctx.add_tools(dup)
assert ctx.tools is not None
assert len(ctx.tools) == 1
def test_add_tools_duplicate_name_different_object_raises():
@tool(name="dup", approval_mode="never_require")
def dup_a(x: int) -> int:
return x
@tool(name="dup", approval_mode="never_require")
def dup_b(x: int) -> int:
return x
ctx = FunctionInvocationContext(function=dup_a, arguments={}, tools=[dup_a])
with pytest.raises(ValueError):
ctx.add_tools(dup_b)
def test_add_tools_batch_with_duplicate_is_atomic():
"""A duplicate-name clash partway through a batch must leave the live list unchanged."""
@tool(name="existing", approval_mode="never_require")
def existing(x: int) -> int:
return x
@tool(name="fresh", approval_mode="never_require")
def fresh(x: int) -> int:
return x
@tool(name="existing", approval_mode="never_require")
def clashing(x: int) -> int:
return x
ctx = FunctionInvocationContext(function=existing, arguments={}, tools=[existing])
with pytest.raises(ValueError):
ctx.add_tools([fresh, clashing])
assert ctx.tools is not None
# The valid "fresh" tool must not have been committed before the clash raised.
assert ctx.tools == [existing]
def test_remove_tools_by_name_and_object():
@tool(name="a", approval_mode="never_require")
def a(x: int) -> int:
return x
@tool(name="b", approval_mode="never_require")
def b(x: int) -> int:
return x
ctx = FunctionInvocationContext(function=a, arguments={}, tools=[a, b])
ctx.remove_tools("a")
assert ctx.tools is not None
assert [t.name for t in ctx.tools] == ["b"]
ctx.remove_tools(b)
assert ctx.tools == []
def test_remove_tools_unknown_name_is_noop():
@tool(name="a", approval_mode="never_require")
def a(x: int) -> int:
return x
ctx = FunctionInvocationContext(function=a, arguments={}, tools=[a])
ctx.remove_tools("nonexistent")
assert ctx.tools is not None
assert [t.name for t in ctx.tools] == ["a"]
def test_progressive_tools_helpers_raise_without_live_tools():
@tool(name="a", approval_mode="never_require")
def a(x: int) -> int:
return x
ctx = FunctionInvocationContext(function=a, arguments={})
assert ctx.tools is None
with pytest.raises(RuntimeError):
ctx.add_tools(a)
with pytest.raises(RuntimeError):
ctx.remove_tools("a")
# endregion

Some files were not shown because too many files have changed in this diff Show More