docs(skills): keep PR reviews focused

[phernandez]

Why

The pull-request workflow correctly requires Codex feedback to be addressed, but it does not
currently require a scope check before turning that feedback into code. That can let severity
labels and speculative edge cases expand a focused fix into unrelated concurrency work,
abstractions, or gold-plating.

The workflow should treat Codex as adversarial review input while preserving human ownership of
the PR's stated outcome and product tradeoffs.

What Changed

• Added an explicit scope-discipline step before acting on review feedback.
• Classified findings as in-scope blockers, sidequests/gold-plating, or fast-follow work.
• Made every regression introduced by the branch an in-scope blocker, even when the PR wording
  omitted the affected behavior.
• Required severity to be justified by reproducible impact and the product contract.
• Added active pushback guidance for speculative or out-of-scope Codex feedback.
• Added a fast-follow issue path for real, material concerns that should not expand the current
  branch.
• Updated the Codex review loop to apply the classification before patching.

Implementation Details

• The skill now anchors scope to the PR's Why, acceptance criteria, and protected behavior.
• Narrow Why or acceptance-criteria wording cannot defer a branch-introduced regression.
• The guidance explicitly avoids generalized locks, leases, retries, migrations, or frameworks
  solely to close theoretical interleavings when eventual consistency is acceptable.
• Out-of-scope feedback is answered with evidence and resolved instead of silently accepted.
• Fast-follow issues require a concrete impact, links to the originating PR/comment, and their own
  acceptance criteria. Issue creation still requires existing user authorization.

Testing

Automated

• just package-check: passed.
  • Claude Code, Codex, shared skill, Hermes, and OpenClaw package validators passed.
  • Hermes: 257 passed, 12 expected integration skips, 87.98% coverage.
  • OpenClaw: 226 passed; typecheck, lint, build, and package dry run passed.

Risks / Follow-ups

• This changes review behavior only; it does not weaken the latest-head CI or Codex approval gate.
• The workflow still fixes concrete in-scope blockers immediately and escalates independently
  critical concerns through a focused fast-follow.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: bf91872ee0

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".