Self-service pentest platform that only scans domains whose ownership is proven, and re-checks that proof at scan time. OWASP, TLS, DNS and port checks + a capped, read-only AI pass.
A free, open-source, verified-domain penetration testing platform. You enter a website, prove you own it, and it runs a real security assessment — OWASP web vulnerabilities, dependency/secret scanning, network/infra exposure, TLS/SSL audit, and (via a bundled Claude Code skill) an AI analysis pass that cross-correlates the results into attack chains a fixed rule engine would miss.
Automated scanning of a website you don't own or control is illegal regardless of intent, even with good motives. So the entire system is built around one non-negotiable gate: no scan ever runs against a domain that hasn't cryptographically proven ownership, re-checked at scan time, not just at signup. Every other safety mechanism below exists to keep this tool from being repurposed into something it isn't.
src/verification/) — DNS TXT record
or a /.well-known/verified-pentest.txt file / meta tag. Re-verified
immediately before every scan, not just once at signup, so a domain that
changed hands or removed the token can't be scanned on stale trust.src/config/safety.ts) — this tool can never become a DoS source.src/utils/ssrfGuard.ts) — every outbound request refuses
to target private/loopback/link-local/cloud-metadata IP ranges.None of this is a checkbox for its own sake — see the check-by-check
rationale inline in the source, especially src/config/safety.ts and
.claude/skills/deep-pentest/SKILL.md.
Zero-install (pure Node, no external binaries):
| Check | What it finds |
|---|---|
headers | Missing CSP/HSTS/X-Frame-Options/etc., server version disclosure |
tls | Cert expiry/trust, deprecated TLS1.0/1.1 support |
dns-config | Missing/weak SPF, missing/monitor-only DMARC |
cookie-flags | Missing Secure/HttpOnly/SameSite on cookies |
csrf | POST forms without an apparent CSRF token (heuristic) |
client-js-libs | Vulnerable client-side JS library versions |
secret-scan | Exposed API keys/secrets in client-served code |
port-scan | Open ports (nmap if installed, built-in TCP-connect fallback otherwise) |
Tool-backed (auto-detected; each check reports skipped with an install
hint if the tool isn't present — nothing fails silently):
| Check | Tool | What it adds |
|---|---|---|
zap-scan | OWASP ZAP | Active web vulnerability scanning (DoS-category rules hard-disabled) |
nuclei-scan | nuclei | Thousands of community CVE/misconfiguration templates |
httpx-fingerprint | httpx | Tech/stack fingerprinting |
subfinder-recon | subfinder | Passive subdomain discovery (reported, never auto-scanned) |
testssl-scan | testssl.sh | Deep TLS/SSL audit (Heartbleed, POODLE, ROBOT, cipher/protocol coverage) |
repo-dependency-scan | gitleaks, osv-scanner, semgrep, trivy | Leaked secrets, vulnerable dependencies, code-level vuln classes, IaC misconfig (only when a repoUrl is supplied) |
npm install
cp .env.example .env # fill in BOOTSTRAP_ADMIN_TOKEN at minimum
npx prisma migrate dev
npm run dev # http://localhost:3000
Optional — installs the tool-backed checks above (no sudo required; nuclei/ httpx/subfinder/trivy/gitleaks/osv-scanner as prebuilt binaries, ZAP + a portable JDK 17, testssl.sh via git clone, semgrep via pipx):
scripts/setup-tools.sh
# 1. Get an API key (gated by BOOTSTRAP_ADMIN_TOKEN in .env — this is a
# self-service security tool, not an open-signup product yet)
curl -X POST localhost:3000/auth/signup \
-H 'content-type: application/json' \
-d '{"email":"[email protected]","bootstrapToken":"<from .env>"}'
# 2. Register a domain -- returns DNS TXT / file / meta-tag instructions
curl -X POST localhost:3000/domains \
-H "authorization: Bearer $API_KEY" -H 'content-type: application/json' \
-d '{"hostname":"yoursite.com"}'
# 3. Add the DNS TXT record (or well-known file / meta tag) to your site,
# then verify:
curl -X POST localhost:3000/domains/$DOMAIN_ID/verify \
-H "authorization: Bearer $API_KEY"
# 4. Scan
curl -X POST localhost:3000/domains/$DOMAIN_ID/scans \
-H "authorization: Bearer $API_KEY" -H 'content-type: application/json' \
-d '{"checkTypes":["headers","tls","dns-config"],"consentStatement":"I own this site"}'
# 5. Poll, then get the report
curl localhost:3000/scans/$SCAN_ID -H "authorization: Bearer $API_KEY"
curl "localhost:3000/scans/$SCAN_ID/report?format=markdown" -H "authorization: Bearer $API_KEY"
Reports render as json (default), markdown, or html. Every report
includes a Check Execution Summary table so "this category found
nothing" is never ambiguous with "this check errored or was skipped."
This repo bundles .claude/skills/deep-pentest/SKILL.md. Open this project
in Claude Code and ask for a deep/AI-assisted pentest on a domain you've
already registered and verified through the API above. The skill:
POST /scans/:id/ai-findings, where it
merges into the same report, clearly badged as AI-CONFIRMED (with
evidence) or AI-HYPOTHESIZED (unconfirmed).The skill file itself carries the hard rules (ownership-verified only, capped/read-only verification, explicit bans on brute-forcing/DDoS/data modification) so they hold regardless of what's asked of it in conversation — read it before assuming what the AI layer will or won't do.
prisma/schema.prisma for what that actually involves), Vitest.src/scanner/orchestrator.ts): p-queue-based,
concurrency-capped, auto-detects the working scheme (http vs https) per
target rather than assuming HTTPS.src/scanner/checks/): all implement the same
CheckModule interface (src/scanner/types.ts) — tool-presence-checked,
skip cleanly if the binary isn't installed, never throw for "found
nothing," only for genuine execution errors.src/utils/networkSetup.ts): in sandboxed/
containerized environments without real IPv6 routing, Node's fetch can
burn 5+ seconds per request on a doomed dual-stack attempt before falling
back. This is forced off by default; revisit if you need to scan
IPv6-only targets.npm test # vitest
npx tsc --noEmit # typecheck
npx prisma migrate dev # after schema changes
MIT — see LICENSE. This is a penetration-testing tool: having the code doesn't grant authorization to scan anything you don't own or have explicit permission to test, on top of what this tool's own verification already enforces.