parler-review

name	parler-review
description	Review a diff, branch, or PR in the Parler Protocol repo against the project review contract (docs/code-review-guidelines.md). Use when asked to review changes, audit a branch, check a PR, or self-review before landing. Produces verified, severity-ranked findings — no style noise, no unverified criticals.
license	Apache-2.0
compatibility	claude-code
allowed-tools	["Read","Grep","Glob","Bash"]

Review a Parler Protocol change

Execute docs/code-review-guidelines.md — read it now if it isn't in context. That document is the contract; this skill is the runbook.

1. Scope: git diff origin/main... plus git status (include working-tree changes). For a PR: gh pr diff <n>.
2. Load: skim "Hard gates" and "Invariants" in docs/engineering-guidelines.md; check tasks/lessons.md for traps in the touched files.
3. Read every changed function in the file, full context — never review from diff hunks alone.
4. Verify every candidate finding: production-reachable (not #[cfg(test)]), concrete failure scenario, cited lines actually read. Drop or downgrade what you can't prove.
5. Run the gates for non-doc changes: scripts/verify.sh, or make ci when more than one crate is touched.
6. Report: severity-ranked findings in the contract's format, then a verdict (approve / approve with nits / needs changes) plus gate results.

This is a read-only pass: never modify files during the review; Bash is for git, gh, and the gates. For a large diff, offload to the code-reviewer agent (.claude/agents/code-reviewer.md) to keep the main context clean.