Is this codebase any good?
Get a third-party answer in five minutes.
Paste a GitHub URL. We read the repo with Claude in an isolated sandbox and email you a grounded report: an executive grade, top risks, and ranked findings with file-and-line citations. Useful before an acquisition, before a hire, before you commit to a vendor — or before you ship. $10, refunded if it fails.
Unescaped separator in regex alternation
A separator string with regex metacharacters silently produces broken slugs. Failures are silent: no exception is thrown.
Grounded findings only
Every finding cites a real file and line range. Ungrounded claims are filtered before delivery — no hallucinated bugs, no made-up CVEs, no advice the model couldn’t back up with code.
Cloned, read, deleted
Your repo is cloned to a sandboxed scratch volume, read once, and wiped within 60 minutes. We email you an HMAC-signed deletion receipt you can verify cryptographically at /verify.
Auto-refund on failure
If the pipeline can’t complete the audit for any reason, your $10 is refunded automatically via Stripe. You never pay for a report that didn’t ship.
Six stages, one quiet hour.
Everything happens between your $10 payment and the report email. You don’t need to keep a tab open.
- 01
Clone
Shallow git clone into a tmpfs scratch dir, read-only mounts elsewhere.
- 02
Filter
Drop vendor, build, lockfiles, minified, generated, binary content. Hard caps applied.
- 03
Triage
Claude Haiku ranks every remaining file by importance. Heuristic fallback if rate-limited.
- 04
Skim
Claude Sonnet reads the ranked files and produces findings, grade, and an exec summary.
- 05
Ground
Each finding's citation is verified against the source. Ungrounded claims dropped.
- 06
Render
Report HTML written, scratch volume wiped, deletion receipt signed and emailed.
Ten dollars. One audit. One email.
We charge once when the audit is queued. If it fails, we refund. No usage tiers, no token meters, no gotchas.