[samirsawarkar]

What and why

Reserves the faultgate-zero name on PyPI before M0 of the build spec (CEO approved 2026-10-21).

• pyproject.toml: hatchling package faultgate-zero 0.0.0, MIT, Python >= 3.11, no dependencies
• src/faultgate_zero/__init__.py: __version__ = "0.0.0" only
• README.md: says this is a placeholder and 0.1.0 is the first usable release
• .github/workflows/ci.yml: build, twine check --strict, install the wheel and import it, on 3.11 / 3.12 / 3.13
• .github/workflows/publish.yml: on a v* tag, checks tag == package version, builds, tests the wheel, publishes through PyPI trusted publishing (environment pypi; pending publisher already added on pypi.org)

Tests

• Local uv build + twine check --strict pass on both sdist and wheel
• CI green on this PR

Review

• Author (agent/model): Claude Code (Claude Opus 5.5)
• Reviewer (different model family) and verdict: Codex, pending

Risks and rollback

Publishing 0.0.0 is public and a PyPI version can't be re-uploaded; it can be yanked. No code ships, so there is nothing to break for users.

🤖 Generated with Claude Code

[samirsawarkar]

AES review: Codex (gpt-6.1-sol) on 63fbd0c: REJECT

• Unapproved dependencies: the build backend (hatchling) and the CI tools build and twine have no OK row in a scout report (.aes/scout/ was empty).
• Verification gaps (not defects): the reviewer ran read-only, so it couldn't run the CI build. CI has since passed on 3.11 / 3.12 / 3.13. The PyPI trusted-publisher setup is outside the repo.
• Passed: version and tag assertions, OIDC id-token: write in environment pypi with no stored secrets, git diff --check, and no .aes/ content committed.

Response

• e8141c8 switches the backend to setuptools>=77, the same as faultgate, so the two repos share one toolchain. The wheel contents are unchanged.
• A focused aes scout is running on setuptools, build, twine, pypa/gh-action-pypi-publish and the actions/* steps. The PR gets a re-review once the scout reports OK rows.

Note: the first review attempt failed before it started, because AES pins gpt-5.5, which this Codex login doesn't support. It was re-run on gpt-6.1-sol.

[samirsawarkar]

AES re-review: Codex (gpt-6.1-sol) on e8141c8: ACCEPT

No blocking findings. The earlier rejection is resolved: .aes/scout/scout-a8ccd56b.md marks setuptools, build, twine, pypa/gh-action-pypi-publish and the actions/* steps OK.

• Metadata, the tag/version guard (accepts v0.0.0, rejects v0.0.1), OIDC publishing with no stored secrets, the placeholder README and the .aes/ ignore all pass.
• The reviewer ran read-only, so it couldn't run the build. CI ran build, twine check --strict and the installed-wheel test on 3.11 / 3.12 / 3.13, all green. The PyPI trusted-publisher setup is outside the repo; the v0.0.0 tag run will show whether it works.