claude-plan-reviewlisted

# Claude Plan Review Independent plan review using Claude Code, pinned to `claude-opus-5` with `--effort max`, `--tools default`, and `--permission-mode auto`. ## Focus Use this for plans, not code diffs. Claude should look for: - spec fidelity gaps: required behavior missing, invented scope, vague acceptance criteria - internal contradictions: thresholds, formulas, naming, ranges, and definitions that disagree - dependency mistakes: milestones ordered incorrectly or pretending to be parallel - data contract gaps: undefined fields, schemas, enums, units, file paths, return shapes - feasibility risks: estimates, rollout assumptions, operational constraints - validation gaps: tests that do not cover the risk-bearing behavior ## Workflow 1. Identify the plan file and any useful context files, such as the spec, upstream model, ADR, schema, or previous roadmap. 2. Run: ```bash python3 ~/.codex/skills/claude-plan-review/scripts/claude_plan_review.py \ <plan_file> \ [context_file ...] ``` 3. Present Claude's findings directly. Do not include Codex's prior reasoning or other reviewer notes in the prompt unless the user explicitly asks for synthesis. The runner streams Claude progress by default, emits heartbeat updates to stderr, and writes durable artifacts under `~/.cache/agent-review-runs/`. Each Claude attempt records `prompt.txt`, `stdout.stream`, `stderr.stream`, `events.jsonl`, `result.md`, and timeout diagnostics. On SIGINT/SIGTERM/timeout, the wrapper terminates