mirror of
https://github.com/microsoft/agent-framework.git
synced 2026-06-16 21:04:09 +08:00
Eduard van Valkenburg
49a6e433a3
* 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>
76 lines
4.8 KiB
Markdown
76 lines
4.8 KiB
Markdown
# Tools
|
|
|
|
Samples that show how to define, configure, and control function tools for an
|
|
agent — from basic declarations to approvals, invocation limits, session
|
|
injection, and dynamic (progressive) tool exposure.
|
|
|
|
## Function tools
|
|
|
|
| File | Demonstrates |
|
|
|------|--------------|
|
|
| [`function_tool_with_explicit_schema.py`](function_tool_with_explicit_schema.py) | Defining a tool with an explicit JSON schema. |
|
|
| [`function_tool_declaration_only.py`](function_tool_declaration_only.py) | A declaration-only tool (schema without a local implementation). |
|
|
| [`function_tool_with_kwargs.py`](function_tool_with_kwargs.py) | Passing extra keyword arguments into a tool. |
|
|
| [`function_tool_from_dict_with_dependency_injection.py`](function_tool_from_dict_with_dependency_injection.py) | Dependency injection into a tool defined from a dict. |
|
|
| [`function_tool_with_session_injection.py`](function_tool_with_session_injection.py) | Injecting the session into a tool. |
|
|
| [`tool_in_class.py`](tool_in_class.py) | Using a method on a class as a tool. |
|
|
| [`agent_as_tool_with_session_propagation.py`](agent_as_tool_with_session_propagation.py) | Exposing an agent as a tool with session propagation. |
|
|
|
|
## Approvals & invocation control
|
|
|
|
| File | Demonstrates |
|
|
|------|--------------|
|
|
| [`function_tool_with_approval.py`](function_tool_with_approval.py) | Requiring human approval before a tool runs. |
|
|
| [`function_tool_with_approval_and_sessions.py`](function_tool_with_approval_and_sessions.py) | Tool approvals combined with sessions. |
|
|
| [`function_invocation_configuration.py`](function_invocation_configuration.py) | Configuring function-invocation settings (e.g. max iterations). |
|
|
| [`control_total_tool_executions.py`](control_total_tool_executions.py) | All the ways to cap how many times tools run. |
|
|
| [`function_tool_with_max_invocations.py`](function_tool_with_max_invocations.py) | Limiting the number of invocations per tool. |
|
|
| [`function_tool_with_max_exceptions.py`](function_tool_with_max_exceptions.py) | Limiting the number of exceptions a tool may raise. |
|
|
| [`function_tool_recover_from_failures.py`](function_tool_recover_from_failures.py) | Returning errors so the agent can recover from tool failures. |
|
|
|
|
## Progressive tool exposure (dynamic loading)
|
|
|
|
| File | Demonstrates |
|
|
|------|--------------|
|
|
| [`dynamic_tool_exposure.py`](dynamic_tool_exposure.py) | A "loader" tool that adds more tools at runtime via `FunctionInvocationContext`. |
|
|
|
|
Frontloading a model with hundreds of tools hurts tool-selection accuracy,
|
|
bloats context, and raises cost. Instead, start with a small set of loader
|
|
tools and let the model pull in more on demand. Inside a tool, the injected
|
|
`ctx: FunctionInvocationContext` exposes a live `ctx.tools` list plus
|
|
`ctx.add_tools(...)` / `ctx.remove_tools(...)` helpers. Tools added or removed
|
|
take effect on the **next iteration** of the function-calling loop.
|
|
|
|
> [!NOTE]
|
|
> Progressive tool exposure applies to the standard function-calling loop. It
|
|
> does **not** apply to CodeAct providers (`agent-framework-monty`,
|
|
> `agent-framework-hyperlight`). In CodeAct the model only sees a single
|
|
> `execute_code` tool, and host tools are exposed *inside the sandbox* as typed
|
|
> Python functions rather than as model tool-schemas. Host tools there are
|
|
> invoked without a `FunctionInvocationContext`, so `ctx.add_tools()` is not
|
|
> available; the helpers fail fast with a clear `RuntimeError` instead of
|
|
> silently doing nothing. To change a CodeAct agent's tool set, use the
|
|
> provider's own `add_tools` / `remove_tool` / `clear_tools` methods (applied
|
|
> between runs). The recommended provider-driven path for Monty and Hyperlight
|
|
> is shown in [`../context_providers/code_act/`](../context_providers/code_act/)
|
|
> ([`code_act.py`](../context_providers/code_act/code_act.py) for Hyperlight,
|
|
> [`monty_code_act.py`](../context_providers/code_act/monty_code_act.py) for
|
|
> Monty).
|
|
|
|
## Local shell & code interpreters
|
|
|
|
| Path | Demonstrates |
|
|
|------|--------------|
|
|
| [`local_shell_with_allowlist.py`](local_shell_with_allowlist.py) | `LocalShellTool` restricted by a strict command allow-list. |
|
|
| [`local_shell_with_environment_provider.py`](local_shell_with_environment_provider.py) | `LocalShellTool` wired with a `ShellEnvironmentProvider`. |
|
|
| [`local_code_interpreter/`](local_code_interpreter/) | Hyperlight-backed sandboxed code interpreter (standalone tool — *extra* pattern). |
|
|
| [`monty_code_interpreter/`](monty_code_interpreter/) | Monty-backed sandboxed code interpreter (standalone tool — *extra* pattern). |
|
|
|
|
> [!TIP]
|
|
> The `local_code_interpreter/` and `monty_code_interpreter/` samples show the
|
|
> standalone-tool wiring and are provided as *extra* reference. For most
|
|
> Monty/Hyperlight use cases the **recommended** path is the provider-driven
|
|
> CodeAct setup in
|
|
> [`../context_providers/code_act/`](../context_providers/code_act/), which adds
|
|
> dynamic tool / capability management.
|