reviewlisted

# Review Review one draft PR produced by the task workflow. Output: PR comments plus a single verdict. Reviewers never push fixes — they report. ## Order matters: tools before opinions 1. **Gather deterministic signal first** (in the PR's worktree or a fresh checkout of the branch): - `vendor/bin/pint --test` — style - `vendor/bin/phpstan analyse` — types/bugs (if installed) - `php artisan test` — behavior - `gh pr diff <number>` — the actual change Any gate failure is an automatic fail verdict; report the output, done. 2. **LLM review only for what tools can't catch**, anchored to diff lines: - does the change do what the task + approved plan say, nothing more? - missing test cases for the changed behavior (not style preferences) - security at trust boundaries: validation, authorization, mass assignment, query injection - N+1 queries, unbounded queries, missing indexes for new query paths - conventions: follows Laravel Boost guidelines, `docs/conventions.md`, and the code-standards skill (cite CS-n in findings) 3. **Skip entirely**: style nits Pint already enforces, hypothetical refactors, scope expansion ("while you're here…"), praise padding. ## Report Post one PR comment via `gh pr comment` (or reply inline where supported): ``` task-agent review — TASK-<n> Gates: pint ✓ / phpstan ✓ / tests ✓ Findings: 1. <file:line> — <problem> — <required fix> (only real findings; empty is fine) Verdict: PASS | FAIL — <one-line rea