← ClaudeAtlas

run-the-gatelisted

Fresh-context spec review before implementation for Rihla money/rules/routing/schema changes. Spawns TWO parallel zero-history Opus subagents per round — a rubric reviewer (7 verification principles) and an orthogonal-axis adversary (off-map regressions, derived surfaces, l10n pairs) — applies the union of P1 findings, and re-runs until both verdicts are clean in the same round. Use before writing code that touches BalanceCalculator/MoneySerializer, firestore.rules or Cloud Functions auth, app_router.dart / deep-links / back-guards, or any field-name/schema change with both a read-path and a write-path. Use when asked to "run the Gate", "gate this spec", or "fresh-context review".
NasserAlbusaidi/ship-gate · ★ 0 · API & Backend · score 72
Install: claude install-skill NasserAlbusaidi/ship-gate
# Run the Gate The in-session author cannot review the in-session author. A spec written this session carries this session's blind spots straight into the implementation. The only fix is a reviewer with **zero session history** — it starts from the code, not from your reasoning, so it structurally catches what your own checklist cannot. This skill replaces `/codex` (retired) with a fresh-context **Opus subagent** as that reviewer. ## When this is mandatory Run BEFORE writing code when the change touches ANY of: - `BalanceCalculator` / money math / `MoneySerializer` - `security/firestore.rules` or Cloud Functions auth/validation - routing — `app_router.dart`, route tree, deep links, back guards - a schema / field-name change with BOTH a read-path and a write-path Outside these (a one-sentence diff, no money/route/schema/rules surface): **skip the Gate, just do it.** Don't burn a review round on a typo. ## The loop 1. **Write the spec to a file** — `docs/plans/<YYYY-MM-DD>-<topic>.md`. While writing, run the 7 verification principles yourself and report results out loud. The embedded check is the reviewer's *rubric*, not a substitute — but a first pass narrows what the reviewer hunts for. 2. **Spawn TWO reviewers in parallel** — both fresh Opus subagents with zero history, launched in a single message so they run concurrently: - **Rubric reviewer** — prompt: the full contents of `reviewer-prompt.md` (this skill's dir) + the absolute path to the spec file. Verifies t