Enforce max_output_length for shell tool outputs

[gustavz]

Motivation

• Prevent overly large shell outputs from overflowing SDK/Responses API limits by making max_output_length actually enforceable.
• Support both structured ShellResult outputs and simple string outputs so tool authors can limit payload size from either side.

Description

• Use the action max_output_length or the executor-returned ShellResult.max_output_length to determine truncation and store the resolved value in the raw payload.
• Add _truncate_shell_outputs to trim structured ShellCommandOutput entries across stdout and stderr, and truncate the rendered output_text to max_output_length.
• Apply truncation for both ShellResult and plain string executor returns, and include the resolved max_output_length in raw_item.
• Modified files: src/agents/_run_impl.py and added a unit test in tests/test_shell_tool.py.

Testing

• Ran the repository verification script bash .codex/skills/verify-changes/scripts/run.sh which runs formatting, linting, type-checking, and the test suite.
• make format and linting passed with no remaining issues.
• mypy completed without errors.
• Test suite outcome: 1186 passed, 4 skipped.

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: be201ee8a8

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 80f39d7fdd

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve action max_output_length=0

If a tool call sets max_output_length to 0 (e.g., to suppress output), _coerce_shell_call currently uses ... or ... to read the field, so 0 is coerced to None. That means requested_max_output_length is None here and this branch won’t truncate at all, even though the caller explicitly asked for zero-length output. This is a regression for the new enforcement path: a model-supplied max_output_length: 0 still leaks full output. Consider preserving zero via an explicit is not None check when parsing the action payload.

Useful? React with 👍 / 👎.

[seratch]

@gustavz This look like an edge case, but can resolve it as well? Also, running codex -m gpt-5.2-codex + /review before pushing again, and resolving reviews if there is anything further would be appreciated