Release 0.19.0

[github-actions]

Release readiness review (v0.18.3 -> TARGET origin/main @ c994c2a)

This is a release readiness report done by $final-release-review skill.

Diff

v0.18.3...c994c2a

Release call:

🟢 GREEN LIGHT TO SHIP No confirmed regression, breaking default behavior, security defect, or packaging failure was found. Complete the standard v0.19.0 release commit and preferably run the live Vercel probe below.

Scope summary:

• 137 files changed (+14,968/-1,629) across 26 commits.
• Key areas: Programmatic Tool Calling, nested-handoff history ownership, RunState schema 1.13, Vercel S3 mounts, sandbox lifecycle, Chat Completions conversion, Realtime redaction, approval parsing, and E2B dependency updates.
• Source: 45 files (+4,381/-298); tests: 28 files (+8,782/-164); docs: 52 files (+1,589/-1,147).

Risk assessment (ordered by impact):

1. Programmatic Tool Calling and nested-handoff persistence

   • Risk: 🟡 MODERATE. New item variants and history ownership touch execution, replay, sessions, handoffs, and serialized resume, but no protocol mismatch or regression was found.
   • Evidence: The implementation handles program, program-owned function_call, and program_output, preserves call_id/caller, copies caller to function outputs, and continues until an assistant message, matching the [official Programmatic Tool Calling contract](https://developers.openai.com/api/docs/guides/tools-programmatic-tool-calling#understand-program-response-items). RunState was correctly bumped to 1.13 while retaining older supported readers. The focused PTC, handoff, session, and RunState tests passed.
   • Files: src/agents/tool.py, src/agents/items.py, src/agents/run_state.py, src/agents/run_internal/turn_resolution.py, src/agents/handoffs/history.py
   • Action: Keep uv run pytest -q tests/test_programmatic_tool_calling.py tests/test_run_state.py tests/test_handoff_history_duplication.py tests/test_extension_filters.py green. Pass condition: all PTC caller-linkage, approval-resume, session, and handoff-history cases pass.

2. Vercel S3 mount lifecycle

   • Risk: 🟡 MODERATE. Provider-side mount ambiguity or cancellation could affect persistence or expose credentials, although the current implementation fails closed and has extensive negative-path coverage.
   • Evidence: Mounts are create-time-only and non-resumable; credentials are removed from serialized state; paths are canonicalized before mounting; tar persistence detaches mounts; ambiguous transitions stop and invalidate the session; cancellation, redaction, hydration overlap, and cleanup cases passed locally.
   • Files: src/agents/extensions/sandbox/vercel/mounts.py, src/agents/extensions/sandbox/vercel/sandbox.py, src/agents/sandbox/util/tar_utils.py, tests/extensions/sandbox/test_vercel.py
   • Action: Run one exact-SHA live Vercel probe covering create, authenticated mount, read/write, persistence, remount, and close. Pass condition: bucket contents are absent from the workspace snapshot, serialized state contains no credentials, and failures expose no secrets.

3. Release version and feature documentation remain release-process steps

   • Risk: 🟡 MODERATE. Publishing the target unchanged would retain version 0.18.3, while the range contains major feature additions.
   • Evidence: pyproject.toml still declares 0.18.3; Programmatic Tool Calling and Vercel mount support add public symbols, but SDK guide coverage for their operational boundaries is limited.
   • Files: pyproject.toml, docs/tools.md, docs/sandbox/guide.md, examples/tools/programmatic_tool_calling.py
   • Action: Use the standard release PR to bump to 0.19.0, update lock metadata, and add release notes covering Responses-only PTC, allowed_callers, Vercel’s create-time/non-resumable boundary, credential opt-in, and snapshot exclusion. Pass condition: built package metadata reports 0.19.0 and the boundaries are documented.

4. E2B optional dependency bump

   • Risk: 🟢 LOW. Users of the optional E2B extra receive newer pinned SDKs, but adapter compatibility remained intact.
   • Evidence: Imports reported e2b==2.31.0 and e2b-code-interpreter==2.8.1; all 71 E2B adapter tests passed.
   • Files: pyproject.toml, uv.lock
   • Action: Keep uv run pytest -q tests/extensions/sandbox/test_e2b.py green. Pass condition: all 71 tests pass.

Notes:

• Remote tags and origin/main were freshly fetched.
• Working tree is clean and aligned with origin/main.
• git diff --check, uv lock --check, public import smoke tests, and the declared minimum openai==2.45.0 PTC type import passed.
• 1,303 focused tests passed. Six UnixLocal tests initially failed because the command sandbox blocked the SDK snapshot directory and macOS sandbox-exec; all six passed when rerun with the required local access.
• No default public behavior was removed. nest_handoff_history remains opt-in, and new public constructor fields are appended or keyword-only.
• No full verification stack was run; per the skill policy, current CI is assumed to cover it.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b3c7255723

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".