This is a submission for the OpenClaw Challenge.
What I Built
I built Nova — an autonomous GitHub PR reviewer
powered by OpenClaw. Every time someone opens a Pull Request,
Nova automatically reads the entire code diff and posts a
structured review comment in under 8 seconds.
No waiting for human reviewers. No rubber stamps.
No secrets leaking into main. Ever.
The core flow:
- Developer opens a PR on GitHub
- GitHub fires a webhook to Nova
- 20 analyzers run in parallel on the diff
- Structured review comment appears in under 8 seconds
- Slack/Telegram notification sent to the team
- Dashboard updates with new risk score and findings
GitHub: https://github.com/noisyboy08/Nova
How I Used OpenClaw
Nova is built as a native OpenClaw skill. The SKILL.md
file defines the entire agent behavior — what triggers it,
what tools it uses, how it analyzes code, and how it learns
from every review.
What OpenClaw adds that a normal bot cannot:
Persistent Memory — OpenClaw remembers every developer
across every PR. After 20+ reviews it builds a pattern
profile per author. It knows which developers forget error
handling. It knows who skips tests. Future reviews are
tailored to that person specifically.
Proactive Notifications — Nova doesn't wait to be asked.
The moment a secret is detected it fires an urgent Telegram
DM to the repo owner. Fully autonomous. No polling.
Skill Composability — Nova chains into other OpenClaw
skills. After a review posts, a Slack notifier skill picks
up the result. A Jira linker connects the PR to the ticket.
Each skill does one thing and hands off cleanly.
Demo
GitHub: https://github.com/noisyboy08/Nova
The 20 analyzers running on every PR:
- 🔑 Secret Scanner — API keys, passwords, tokens
- 📦 CVE Checker — Vulnerable packages via OSV.dev
- 🛡️ OWASP Top 10 — All 10 categories per PR
- 📊 Risk Score — 0-100 with colored GitHub label
- 🔧 Auto-Fix — Writes the fix as inline suggestion
- 🧪 Test Generator — Generates missing unit tests
- 🏗️ Architecture — Blast radius for core file changes
- ⚖️ License Checker — Flags GPL in commercial projects
- 💳 Sensitive Data — Emails, SSNs in test data
- 👥 Auto-Assign — Git blame based reviewer suggestions
- 🎫 Jira Linker — Connects PR to ticket from branch name
- 📝 Changelog — Auto-commits entry on merge
- 🏆 Leaderboard — Per-developer metrics and streaks
- 🎭 Personalities — Strict / Mentor / Fast / Senior modes
- 📱 Notifications — Slack, Telegram, Discord cards
- 🔍 Pattern Learning — Remembers each author's mistakes
- 💡 Praise — Calls out genuinely elegant code
- 📏 Size Enforcer — Suggests how to split huge PRs
- 🌍 Multi-Language — Python/JS/Go/Java specific rules
- 📜 Git Context — Last 5 commits on changed files
Dashboard: Dark glassmorphism design with animated
risk gauge, team leaderboard, security timeline,
and real-time review feed.
What I Learned
1. OpenClaw's SKILL.md is the most underrated
feature in personal AI right now.
One Markdown file replaced what would have been
hundreds of lines of orchestration code.
2. Parallel execution changes everything.
20 analyzers sequentially = 40 seconds.
Same 20 with Promise.allSettled() in parallel = 8 seconds.
Architecture matters more than features.
3. The webhook pipeline is harder than the AI.
Verifying HMAC signatures, responding within 2 seconds,
processing async — that plumbing took longer than
building all 20 analyzers combined.
4. OpenClaw makes AI proactive not reactive.
Every other AI tool waits for you to ask. OpenClaw
watches your workflow and acts without being prompted.
That shift is what makes it fundamentally different
from everything else.
5. Personal AI is about to eat developer tooling.
Nova does what CodeClimate and Snyk do — but
self-hosted, free, and customizable in a single
Markdown file. A solo developer can build this
in 10 days with OpenClaw.
ClawCon Michigan
I did not attend ClawCon Michigan but the energy of
the OpenClaw community building in public inspired
every decision in this project.
The lobster stays. 🦞