skills/review-all-prs/SKILL.md
Dispatch parallel agents to address review comments on all open PRs. Each agent fixes code, replies inline, resolves threads, and reports what needs human input. Use when user says '/review-all-prs', 'fix all PR reviews', 'address all reviews', or when /scrum shows PRs with pending reviews.
npx skillsauth add razbakov/skills review-all-prsInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Dispatch one agent per PR to address review comments. Agents fix what they can autonomously. Human gets a summary of what needs decisions.
For each project in ~/Projects/ that has open PRs:
cd ~/Projects/PROJECT
REPO=$(gh repo view --json nameWithOwner --jq '.nameWithOwner')
gh pr list --state open --json number,title,url,headRefName
For each PR, check for unresolved threads:
gh api graphql -f query='
query {
repository(owner: "OWNER", name: "REPO") {
pullRequest(number: NUMBER) {
reviewThreads(first: 50) {
nodes { isResolved }
}
}
}
}' --jq '[.data.repository.pullRequest.reviewThreads.nodes[] | select(.isResolved == false)] | length'
Skip PRs with 0 unresolved threads.
For each PR with unresolved threads, dispatch using the /inbox pattern:
pr-review-${PR_NUMBER}~/Tasks/${PROJECT}-pr-review-${PR_NUMBER}The worktree setup is different from normal /inbox — we need the PR branch, not a new branch:
TASK_DIR=~/Tasks/${PROJECT}-pr-review-${PR_NUMBER}
git -C ~/Projects/${PROJECT} worktree add ${TASK_DIR} ${PR_BRANCH}
Each agent gets this prompt:
# Address PR review comments for PR #NUMBER
**Repo:** OWNER/REPO
**PR:** #NUMBER — TITLE
**Branch:** BRANCH
**URL:** PR_URL
## Instructions
You are addressing review comments on this PR. Follow the pr-review-responder workflow:
1. Fetch all inline review comments:
`gh api repos/OWNER/REPO/pulls/NUMBER/comments --jq '.[] | {id, path, line, body}'`
2. Categorize each comment:
- **Code fix needed** → Fix the code, commit with SHA reference
- **Acknowledged (missing feature)** → Create a GitHub issue, reply with link
- **Acknowledged (product decision)** → DO NOT fix or create issue. Write to NEEDS_INPUT file instead.
- **By design** → Reply with reasoning
- **Already fixed** → Reply noting the commit
3. Fix all code issues, then verify:
```bash
bun run build 2>&1 || npm run build 2>&1 || true
Commit and push:
git add -A && git commit -m "Address PR #NUMBER review feedback" && git push
Reply to each comment inline:
gh api -X POST repos/OWNER/REPO/pulls/NUMBER/comments -f body="Fixed in SHA" -F in_reply_to=COMMENT_ID
Resolve threads via GraphQL (get thread IDs first, then resolveReviewThread mutation)
For items needing human input, append to ${TASK_DIR}/needs-input.md:
## PR #NUMBER — TITLE
- [ ] Comment ID XXXX: <summary of what needs deciding>
### 4. Launch all agents in parallel
Use the standard `/inbox` tmux launch pattern for each PR. Session name: `wf-pr-review-${PR_NUMBER}`.
### 5. Report dispatch summary
After dispatching all agents, output:
Dispatched N agents to address PR reviews:
Run /scrum to check progress.
## Collecting Results
When agents finish (check via `/scrum`), collect results:
```bash
# Check for needs-input files across all review agents
for dir in ~/Tasks/*-pr-review-*/; do
if [ -f "$dir/needs-input.md" ]; then
echo "=== $(basename $dir) ==="
cat "$dir/needs-input.md"
fi
done
Present to user:
## Review Results
### Fully resolved (ready to test)
- PR #23: Add hero image — all 3 threads fixed and resolved
- PR #24: Wire useDinners — all 2 threads fixed and resolved
### Needs your input
- PR #26: Signup form
- [ ] Should email validation use regex or API check?
- [ ] Password requirements: min 8 chars or min 12?
### Failed (check logs)
- PR #27: Agent crashed — `tail ~/Tasks/project-pr-review-27/agent.log`
tools
--- name: handoff description: Get an agent past a browser/UI wall it can't (or must not) cross on its own — a login-gated dashboard, a CAPTCHA, a 2FA prompt, an API that keeps rejecting the write, or an irreversible click that policy says a human must make. This skill is an ESCALATION LADDER, not a first move: it tells you to try the automated browser surfaces FIRST (Chrome-in-Claude, computer-use, an autonomous browser sub-agent) and only fall back to the Handoff app — a wrapper browser that h
documentation
Summarize one or more YouTube videos from their links. Use this whenever the user pastes a youtube.com or youtu.be URL (or several) and wants to know what it's about — phrasings like "summarize to telegram", "tldr these videos", "what do these say", "summary of this talk", or just dropping links with no instruction at all. Fetches each video's real transcript via yt-dlp (not the page text, which never contains the transcript), cleans the captions, and writes a per-video summary. Default delivery is Telegram; honor any other surface the user names ("to my notes", "just here in chat", "email it"). Trigger even when the user only pastes bare links — bare YouTube links almost always mean "tell me what's in these".
data-ai
Daily Digest — Chief-of-Staff role consolidates the six top-managers into one Telegram message to the Commander, instead of six. Implements the protocol from agent-proactivity.md.
development
Seed a new or empty Instagram account with a 9-post grid (3×3) so the profile looks established the moment a new visitor lands. Designed for festivals, new businesses, product launches, conferences, communities — any time an empty IG profile would hurt conversion from external traffic (QR scans, flyer drops, cross-promo). Generates assets via /image-from-gemini (per content-publishing rules — never HTML), writes captions with hashtag sets, and outputs a posting order + cadence plan. Trigger generously: phrases like '9 posts for instagram', 'fill my IG', 'starter grid', 'launch grid', 'instagram seed', '9-post grid', 'IG account not to look empty', 'first instagram posts', 'feed bootstrap', '3x3 grid', 'instagram launch content'. Even if the user mentions only one piece (just the images, just the captions, just the order), use this skill — the grid only works as an integrated bundle.