I've put together a working triage skill that I think could be a strong starting point or complement to whatever you're building. Here's what it does and how I'd like to contribute.
A complete, ready-to-use OpenClaw skill for triaging PRs and issues at scale. It's structured to handle the volume of the OpenClaw repo specifically.
| Feature | Description |
|---|---|
| Single PR Triage | Classifies type (bug-fix, feature, refactor, etc.), estimates complexity & risk, checks CI/review status, flags security-sensitive changes, suggests reviewers |
| Single Issue Triage | Classifies priority (P0–P3), detects incomplete issues, finds potential duplicates, checks for linked PRs |
| Batch Mode | Processes up to 200 PRs/issues at once with filtering by state, recency, and labels — designed for the 3.6k+ backlog |
| PR Dashboard | Generates an action-prioritized overview: 🔴 Immediate, 🟡 Needs Review, 🟢 Ready to Merge, ⚪ Stale, 🔵 Deps |
| 🔍 Duplicate & Plagiarism Detection | Compares PRs by file overlap, submission timing, authorship, title similarity, and branch names to flag copied or duplicate work |
| 🚧 CI Blocker Analysis | Groups CI failures by check name and ranks by "blast radius" — how many PRs each fix unblocks. Identifies CI-fix PRs as priority merges |
| 👤 Contributor Insights | First-time contributor detection, merge rate history, contribution tier, bot detection, and adaptive review recommendations |
openclaw-triage-skill/
├── SKILL.md # Skill definition with classification rules
├── scripts/
│ ├── triage-pr.sh # Fetches PR metadata via gh CLI
│ ├── triage-issue.sh # Fetches issue metadata + duplicate detection
│ ├── batch-triage.sh # Bulk fetching with filters
│ ├── detect-duplicates.sh # 🔍 Duplicate/plagiarism detection across PRs
│ ├── ci-blockers.sh # 🚧 CI failure grouping & blast radius analysis
│ └── contributor-profile.sh # 👤 Author history, merge rate, tier classification
└── templates/
└── triage-report.md # Structured output template
All scripts use the gh CLI (no API tokens to manage separately), work cross-platform (macOS + Linux), and respect API rate limits with configurable batch sizes.
## PR #1234 — Fix authentication bypass in OAuth flow
| Field | Value |
|-------------|----------------------------------|
| Type | 🐛 bug-fix |
| Complexity | small (23 lines) |
| Risk | 🔴 critical (touches auth code) |
| CI | ✅ passing |
| Review | approved |
| Staleness | 2 days |
Summary: Fixes a logic error in the OAuth callback handler that
could allow bypass of email verification under specific conditions.
→ Recommended action: **MERGE** (security fix, approved, CI green)
Here's my concrete plan for tackling the backlog if I join the team:
- 🔍 Flag duplicate/stolen PRs: Run duplicate detection across the open PR queue to identify copied work — a known pain point for maintainers dealing with credit disputes and review waste
- Close abandoned PRs: Batch-filter for PRs with 90+ days of inactivity, no reviews, and failing CI → generate a "stale PR" report for maintainer approval before closing
- Merge dependency bumps: Batch-identify Dependabot/Renovate PRs with passing CI → flag for auto-merge
- Label the unlabeled: Run triage on all unlabeled PRs to auto-suggest labels
- 🚧 Unblock CI bottlenecks: Identify which CI failures block the most PRs and prioritize fixing them — a single lint fix can unblock 20+ PRs
- Daily triage of new PRs: Set up a recurring job that triages all PRs opened in the last 24 hours and posts summaries to the team channel
- 👤 Contributor-aware reviews: Use contributor profiles to fast-track trusted contributors and provide onboarding guidance for first-timers
- Priority queue: Generate weekly "action needed" dashboards sorted by risk and impact
- Reviewer routing: Use git blame + CODEOWNERS to suggest the right reviewer for each PR area
- GitHub Action integration: Plug the skill into
openclaw-github-appso every new PR gets auto-triaged on open - Metrics tracking: Track triage throughput (PRs triaged/day, median time-to-first-review) to measure impact
- Community-facing dashboard: A public "PR status" page showing the health of the review queue
- Aligning with your existing triage work — Happy to adapt the skill to fit whatever architecture you're building
- Access and permissions — What level of repo access does the triage team need?
- Communication — Which channel does the team use for triage coordination? (Slack? Telegram? Discord?)
The skill is ready to test — you can drop the openclaw-triage-skill/ folder into your skills directory and try triage PR #<any-number>. Looking forward to working together on this!
— Dinakar