VIBE CODED.
VIBE HARDENED.®
One command. No config. Scans code AND agent skill files generated by Cursor, Claude Code, OpenClaw, Hermes, v0, Lovable, Bolt — knows what each tends to get wrong.
npx vibe-hardening scan● 74 CODE RULES
● 65 AGENT RULES
● 9 LIVE VERIFIERS
DRIZZLE · CONVEX · VERCEL · NETLIFY
might look the same.
vibe-hardening scanMain command. Scores your repo 0–100 with A–F grade. 74 rules covering hardcoded keys, SQL injection, missing auth on routes, CORS, Supabase RLS, eval(req.body), localStorage tokens, weak bcrypt rounds, and packages LLMs hallucinate.
scan --changed-only [ref]Scan only files in git diff. Without a ref: vs HEAD. With a ref like origin/main: 3-dot diff for PR-mode CI scans. 10× faster on large repos.
scan --verify --ownHits each leaked key against the real provider API. 9 providers. Tells you which are still live vs. revoked. --own is a seatbelt that refuses to probe keys you have not claimed.
scan --suggest-fixPrints copy-paste-able diffs that swap inline keys for process.env.X plus an .env.example stub. Console-only. Never modifies your files.
scan --roastBrutalist mode. Neutral rule messages become dry one-liners. Console only — JSON / HTML output stays professional for CI artifacts.
vh explain <rule-id>Detailed docs for any rule: severity, what it detects, why it matters, how to fix. Covers every shipped rule ID. Docs in your terminal — no browser needed.
vh badgeOutputs an SVG you can embed in your README to show the repo current security score. Live-updating when paired with a scheduled CI run.
AI-aware rules
Hand-tuned for v0 exports, Cursor loops, Lovable scaffolds, Bolt scaffolds, and Claude Code diffs. Knows what each tends to miss — and won't complain about what they get right.
Platform fingerprint
Detects Next.js, Supabase, tRPC, Prisma, Drizzle, Convex, Vercel, Netlify. Runs only the rules that apply to your stack, so scans stay under 5 seconds on average repos.
Live secret verification
Finds candidate keys in your git history, probes their provider endpoints, and tells you which ones are still live. Not a grep — a phone call.
Agent skill scanner
New in 0.4.0. Statically scans skill files for Cursor, Claude Code, OpenClaw, Hermes, Gemini CLI, Goose, and 4 other agent platforms. Catches hardcoded keys, prompt injection, dangerous shell, MCP misconfigs — before the agent loads them.
Free, forever.
Scan any repo. Unlimited runs. Your code stays on your machine — only opt-in anonymous stats (rule-IDs that fired, never paths or content).
npx vibe-hardening scanSHIP
HARDENED.
One email on launch day, 2026-05-13, 14:00 UTC. No marketing.