qa

QA

Standalone QA run for any PR. Boots the local environment, validates every acceptance criterion, and produces a test report. Posting to GitHub is your choice — you are prompted at the end.

Step 1 — Load config

Read project config from the orchestrator's ## Project Config block:

ORCHESTRATOR=".claude/skills/orchestrator/SKILL.md"
REPO=$(grep '^REPO=' "$ORCHESTRATOR" | cut -d= -f2)
TEMP_ROOT=$(grep '^TEMP_ROOT=' "$ORCHESTRATOR" | cut -d= -f2)
BOOT_CMD=$(grep '^BOOT_CMD=' "$ORCHESTRATOR" | cut -d= -f2-)
LOCAL_URL=$(grep '^LOCAL_URL=' "$ORCHESTRATOR" | cut -d= -f2)

Step 2 — Resolve the PR

Use $ARGUMENTS as the PR number or URL. If empty, resolve from the current branch:

gh pr list --head "$(git branch --show-current)" --json number,url -q '.[0] | "\(.number) \(.url)"'

If no PR is found, tell the user and stop.

Get the base branch:

gh pr view <PR_NUMBER> --json baseRefName -q .baseRefName

Step 3 — Invoke the qa-engineer agent

Invoke the qa-engineer sub-agent with:

• PR number and PR URL
• Base branch from Step 2
• Runtime values: TEMP_ROOT={TEMP_ROOT}, REPO={REPO}, E2E_BOOT={BOOT_CMD}, E2E_URL={LOCAL_URL}

STANDALONE MODE — two differences from the normal pipeline run:

1. Skip Step 6 (posting the PR comment). Instead, output the full QA report as formatted Markdown in your response, in a section titled ## QA Report. Use the same format the pipeline would post (including the <!-- ai-pipeline:qa-report --> marker).
2. Skip the StructuredOutput JSON return. Output a short human-readable summary instead: overall result, pass/fail per criterion, and any blockers.

All other steps run normally — the environment is booted ({BOOT_CMD}), acceptance criteria are tested, and the full validation is performed.

Step 4 — Offer to post

After the agent responds, display its ## QA Report and ask:

Post this QA report to PR #<PR_NUMBER>? Reply yes to post, no to finish here.

If yes — post with dedup: check for an existing <!-- ai-pipeline:qa-report --> comment, update it with PATCH if found, otherwise create a new comment.

If no — confirm the QA run is complete and finish.