Authorized, non-destructive website pentests — for sites you have proven you own.
auto.hacker runs a security scan (missing headers, reflected-XSS markers, CORS, cookie flags, CSP, directory listing, error disclosure, sensitive paths, unsafe HTTP methods, SQLi/traversal probes) against a website. It will not touch a host until the signed-in user has proven they control that exact host, and the proof is re-checked at scan launch and again in the worker right before the first request. The engine is ThreatCrush.
Live: https://auto.hacker.pit.world
No scan reaches the engine until a per-user-per-host proof is found live. Token:
HMAC(secret, "<user_id>:<host>"). Publish one:
- DNS TXT —
_autohacker.<host>=autohacker-verify=<token> - File —
https://<host>/.well-known/autohacker-<token>.txtcontaining the token - Meta —
<meta name="autohacker-verify" content="<token>">on the homepage
A host you cannot prove you own is refused. Every verification appends an immutable authorization log row. See the manifesto.
- Web (PWA): add target, proof + verify, launch scan, live SSE run stream, history, CSV/JSON export, account/profile.
- API:
GET/POST /api/v1/targets,/targets/:id/verify,POST /api/v1/runs,/runs/:id,/runs/:id/stream(SSE),/runs/:id/export.{json,csv},GET /api/v1/me/profile(401 logged out). Bearerah_…. - CLI:
autohacker add|proof|verify|scan|runs|report|targets. - MCP:
autohacker mcp(add_target, get_proof, verify, scan, runs, report, targets). - TUI:
autohacker tui.
A run streams live over SSE. A verified host can be subscribed to periodic (daily/weekly/monthly) scans; the worker runs them and emails a summary (Resend). Scheduled scans are the paid tier (CoinPay); on-demand scans of your own verified sites are free.
Authorized testing of your own verified properties, non-destructive checks only. Not a substitute for a manual penetration test.
bun install
AUTOHACKER_KEY=$(head -c32 /dev/urandom | od -An -tx1 | tr -d ' \n') bun src/server.ts
bun test # the ownership gate is the first and most-tested thing here
MIT.