mirror of
https://github.com/pchuan98/codex.git
synced 2026-07-01 00:31:56 +08:00
Ahmed Ibrahim
b1cbf622ad
## Why
The Python SDK currently exposes sandbox selection differently depending
on where it is used: thread lifecycle methods accept `SandboxMode`,
while turns accept the lower-level `SandboxPolicy` shape. For the common
case of choosing an access level, that leaks app-server wire details
into otherwise straightforward SDK usage.
This makes the common path explicit and discoverable: callers choose a
named sandbox preset once, using the same keyword on threads and turns.
The preset name `workspace_write` also makes the granted capability
clear at the callsite.
## What changed
- Added a root-level `Sandbox` enum with documented presets:
- `Sandbox.read_only`: read files without allowing writes.
- `Sandbox.workspace_write`: the normal default for projects with a
recorded trust decision; read files and write inside the workspace and
configured writable roots.
- `Sandbox.full_access`: run without filesystem access restrictions.
- Documented that omitting `sandbox=` delegates to app-server's
configured default, while explicit turn overrides remain sticky for
subsequent turns.
- Updated sync and async thread lifecycle and turn APIs to consistently
accept `sandbox=Sandbox...`, translating to the existing app-server
thread and turn representations internally.
- Updated the public API artifact generator so regenerated SDK wrappers
retain the friendly enum shape.
- Replaced low-level policy construction in Python docs, examples, and
the walkthrough notebook with the preset API.
- Added focused coverage for root exports, method signatures,
preset-to-wire mapping, and rejection of raw string sandbox inputs.
## API impact
High-level turn calls now use `sandbox=` instead of `sandbox_policy=`:
```python
from openai_codex import Codex, Sandbox
with Codex() as codex:
thread = codex.thread_start(sandbox=Sandbox.workspace_write)
result = thread.run("Review the diff only.", sandbox=Sandbox.read_only)
```
`thread_start(...)` already defaults to `ApprovalMode.auto_review`, so
normal writable usage is concise:
```python
with Codex() as codex:
thread = codex.thread_start(sandbox=Sandbox.workspace_write)
thread.run("Update the files in this workspace.")
```
With that combination, edits inside `cwd` and configured writable roots
run within the workspace-write sandbox. Operations that require
approval, such as edits outside those roots, are routed through auto
review. When `sandbox=` is omitted, app-server resolves its configured
default. A sandbox supplied to `run(...)` or `turn(...)` applies to that
turn and subsequent turns.
## Test coverage
- `sdk/python/tests/test_public_api_signatures.py` covers the public
export and parameter names, including the default approval mode.
- `sdk/python/tests/test_public_api_runtime_behavior.py` covers preset
mappings to the existing wire types and raw string rejection.
110 lines
4.0 KiB
Markdown
110 lines
4.0 KiB
Markdown
# FAQ
|
|
|
|
## Thread vs turn
|
|
|
|
- A `Thread` is conversation state.
|
|
- A `Turn` is one model execution inside that thread.
|
|
- Multi-turn chat means multiple turns on the same `Thread`.
|
|
|
|
## `run()` vs `stream()`
|
|
|
|
- `Thread.run(...)` starts a turn and returns `TurnResult`.
|
|
- `TurnHandle.run()` / `AsyncTurnHandle.run()` consumes events for an existing turn handle and returns the same `TurnResult` shape.
|
|
- `TurnHandle.stream()` / `AsyncTurnHandle.stream()` yields raw notifications (`Notification`) so you can react event-by-event.
|
|
|
|
Choose `run()` for most apps. Choose `stream()` for progress UIs, custom timeout logic, or custom parsing.
|
|
|
|
## Sync vs async clients
|
|
|
|
- `Codex` is the sync public API.
|
|
- `AsyncCodex` is an async replica of the same public API shape.
|
|
- Prefer `async with AsyncCodex()` for async code. It is the standard path for
|
|
explicit startup/shutdown, and `AsyncCodex` initializes lazily on context
|
|
entry or first awaited API use.
|
|
|
|
If your app is not already async, stay with `Codex`.
|
|
|
|
## How do I log in?
|
|
|
|
- `login_api_key(...)` authenticates immediately with an API key.
|
|
- `login_chatgpt()` starts browser login and returns a handle with `auth_url`.
|
|
- `login_chatgpt_device_code()` starts device-code login and returns a handle
|
|
with `verification_url` and `user_code`.
|
|
- Interactive handles expose `wait()` for the matching
|
|
`account/login/completed` notification and `cancel()` to stop that attempt.
|
|
- `account()` reads the current account state, and `logout()` clears it.
|
|
|
|
## Public kwargs are snake_case
|
|
|
|
Public API keyword names are snake_case. The SDK still maps them to wire camelCase under the hood.
|
|
|
|
If you are migrating older code, update these names:
|
|
|
|
- `approvalPolicy` -> `approval_policy`
|
|
- `baseInstructions` -> `base_instructions`
|
|
- `developerInstructions` -> `developer_instructions`
|
|
- `modelProvider` -> `model_provider`
|
|
- `modelProviders` -> `model_providers`
|
|
- `sortKey` -> `sort_key`
|
|
- `sourceKinds` -> `source_kinds`
|
|
- `outputSchema` -> `output_schema`
|
|
|
|
## How do I choose sandbox access?
|
|
|
|
Use the same `sandbox=` keyword for threads and turns:
|
|
|
|
```python
|
|
from openai_codex import Sandbox
|
|
|
|
thread = codex.thread_start(sandbox=Sandbox.workspace_write)
|
|
result = thread.run("Review only.", sandbox=Sandbox.read_only)
|
|
```
|
|
|
|
The presets are:
|
|
|
|
- `Sandbox.read_only`: read files without allowing writes.
|
|
- `Sandbox.workspace_write`: the normal default for projects with a recorded trust decision; read files and write inside the workspace and configured writable roots.
|
|
- `Sandbox.full_access`: run without filesystem access restrictions.
|
|
|
|
When `sandbox=` is omitted, app-server uses its configured default. A turn
|
|
sandbox override applies to that turn and subsequent turns.
|
|
|
|
## Why only `thread_start(...)` and `thread_resume(...)`?
|
|
|
|
The public API keeps only explicit lifecycle calls:
|
|
|
|
- `thread_start(...)` to create new threads
|
|
- `thread_resume(thread_id, ...)` to continue existing threads
|
|
|
|
This avoids duplicate ways to do the same operation and keeps behavior explicit.
|
|
|
|
## Why does constructor fail?
|
|
|
|
`Codex()` is eager: it starts transport and calls `initialize` in `__init__`.
|
|
|
|
Common causes:
|
|
|
|
- published runtime package (`openai-codex-cli-bin`) is not installed
|
|
- local `codex_bin` override points to a missing file
|
|
- app-server version older than the SDK schema
|
|
|
|
## Why does a turn "hang"?
|
|
|
|
A turn is complete only when `turn/completed` arrives for that turn ID.
|
|
|
|
- `run()` waits for this automatically.
|
|
- With `stream()`, keep consuming notifications until completion.
|
|
|
|
## How do I retry safely?
|
|
|
|
Use `retry_on_overload(...)` for transient overload failures (`ServerBusyError`).
|
|
|
|
Do not blindly retry all errors. For `InvalidParamsError` or `MethodNotFoundError`, fix inputs or update the runtime/schema version instead.
|
|
|
|
## Common pitfalls
|
|
|
|
- Starting a new thread for every prompt when you wanted continuity.
|
|
- Forgetting to `close()` (or not using context managers).
|
|
- Reading `Turn.items` from live start/completed payloads instead of using `TurnResult.items`.
|
|
- Mixing SDK input classes with raw dicts incorrectly.
|