PR Check Skill

You are a PROACTIVE GitHub Actions assistant. After EVERY git push, you MUST automatically verify all GitHub Actions workflows are successful. If any fail, read error logs and resolve issues.

Run the Check Script

The script accepts optional args that can be passed via the skill invocation (e.g. /check-pr links, /check-pr comments, /check-pr links comments).

Map skill args to script flags:

# Default — poll CI, report result
.claude/skills/check-pr/run-workflow-check.sh

# With per-test deep links
.claude/skills/check-pr/run-workflow-check.sh --links

# Comments only
.claude/skills/check-pr/run-workflow-check.sh --comments-only

# Comments + links (can combine)
.claude/skills/check-pr/run-workflow-check.sh --comments-only --links

When user passes args, translate them:

• links → append --links
• comments → append --comments-only

This script will automatically:

1. Install gh CLI if not available
2. Check authentication
3. Get current branch
4. Poll workflow status every 10 seconds until complete
5. Report SUCCESS or FAILURE with full logs

Deciding When to Use --links

Pass --links automatically (without the user asking) when CI completes with test failures in the E2E workflow. Do NOT pass --links on pure success runs unless the user explicitly requests it.

Filtering Which Links to Show

When --links output is available, do not show all test links. Show only the relevant subset:

1. Failing tests — always show links for any test with status unexpected or failed
2. Tests related to changed files — determine which spec files correspond to files modified in this PR/push (e.g., changes in server/gameplay/knight.py → show links from e2e/basic.spec.js tests that cover the knight), then show links for those tests regardless of pass/fail

Omit all other passing tests that are unrelated to the current change set. This keeps the output focused and actionable.

Output Format for Test Links

Always present test links in a plain-text format suitable for copying to messaging apps (e.g. WhatsApp). Use this exact format:

✅ PR #<number> — All checks passed

🧪 <Feature name> E2E Tests:

1. <test title>
<full URL with #?testId=...>

2. <test title>
<full URL with #?testId=...>

• Use ✅ for success, ❌ for failure
• Number each test
• Put the full URL on its own line (no markdown link syntax)
• Group tests by spec file with a header if multiple spec files are shown

On Failure

1. Get logs from failed runs:

``bash gh run view RUN_ID --repo OWNER/REPO --log-failed ``

1. Analyze, fix, commit, push, and re-run the script until workflow passes.

Address PR Review Comments

The script displays unresolved comments with their IDs. Address ONLY unresolved comments.

Option A: Fix the issue

1. Fix and commit
2. Push changes
3. Reply within the comment thread: [Agent] Fixed - <what was fixed>

Option B: Ask for clarification

• Reply within the comment thread: [Agent] Question: <your question>

Reply within comment thread (REQUIRED):

Use the /replies endpoint to respond within the same comment thread (not as a separate comment):

gh api repos/OWNER/REPO/pulls/PR_NUMBER/comments/COMMENT_ID/replies \
  -f body="[Agent] Fixed - <description of what was fixed>"

Example:

# Reply to comment ID COMMENT_ID on PR #PR_NUMBER
gh api repos/OWNER/REPO/pulls/PR_NUMBER/comments/COMMENT_ID/replies \
  -f body="[Agent] Fixed - <description of what was fixed>"

IMPORTANT:

• Always prefix responses with [Agent] to identify agent-generated replies
• Use the /replies endpoint to keep responses in the comment thread
• Only respond to unresolved comments
• Be concise in responses

Be PROACTIVE: Check and poll workflows after every push!