← ClaudeAtlas

qa-gatelisted

Run QA gate validation at a layer boundary. Spawns QA subagent, validates results (fabrication detection, premature criteria, infrastructure allowlist), routes failures to qa-fixer. Re-invoke at the start of each new layer.
willietran/autoboard · ★ 3 · Testing & QA · score 47
Install: claude install-skill willietran/autoboard
# QA Gate Run acceptance testing and regression checks at a layer boundary. The orchestrator spawns a QA subagent (Agent tool), then exercises its most critical judgment: validating the QA agent's claims before routing the result. **Prerequisites:** - A COHERENCE-REPORT must exist for this layer (from the coherence-audit skill). If you do not have one, STOP and run the coherence audit first. The QA brief requires the COHERENCE-REPORT. - All sessions in this layer must be merged to the feature branch. --- ## Tracking: Move to Verifying **First action** before any QA work. Read `docs/autoboard/{slug}/github-tracking.md` to get the QA gate issue number and item ID for this layer. If tracking is active: `move-ticket(qa-gate, Verifying)` --- ## Pre-QA Environment Sync Run the manifest's `setup` command now (if configured) to ensure the environment is current with all merged code from this layer. This is intentionally redundant with step 4b in the run skeleton — QA must test against a fresh environment, and the setup command is idempotent. If the setup command fails, diagnose and fix before spawning the QA agent. A QA gate against a broken environment produces false failures that waste fixer attempts. --- ## Spawn QA Subagent **QA Prompt Integrity:** The Agent prompt below is a FIXED TEMPLATE. Fill in the `{placeholder}` fields with data from the manifest, config, and prior steps. Do NOT: - Add skip instructions beyond what's in the manifest's `expected-skips` list - T