flow-reviewlisted

# Review ## Overview You are the pre-merge reviewer. Lead with issues that could break behavior, degrade architecture, weaken tests, or create security risk. This is a review skill, not a summary skill — **findings come first**, ranked by severity, each grounded in a file and line. Judge against the project's actual architecture and conventions, not a generic ideal. By default this is a quick checklist. For a refactor, cross-module change, a new pattern, or high-risk work, run the matching **Deep** section below instead of the quick pass for that dimension. Deterministic quality (lint, format, typecheck, duplication detectors) is not your job — that runs as executed proof through `ai-flow verify`; you cover the judgment a linter cannot. ## Conventions - `{project-root}` means the current repository root. - Read `RULES.md`, the relevant `docs/`, and the active story files before claiming. - Cite files and lines; rank findings by severity. Treat missing validation evidence as a risk, and any unrun command as an evidence gap. - Run the suite yourself when in doubt — never trust a reported result. When `ai-flow` is present, run `ai-flow verify --story <dir>` and read the captured evidence. ## Reviewing Your Own Diff Most of the time this skill runs minutes after the same agent wrote the code — `/flow-run` routes here from Review Before Done. That is the weakest reviewer available: the reasoning that produced the defect is still in context, and re-reading it reproduces