Security audits for Solana programs, with a record on-chain.

The repository is sized through the GitHub API, without cloning. You see the expected model cost for each supported model before anything runs. The figure is an upper bound fitted on our own audits, with a 1.4× safety factor; the calibration data is in the repository.

You create a dedicated key in an Anthropic workspace with a spend limit (the page tells you the number) and paste it. We verify it with one free API call, encrypt it, and wipe it the moment the run starts. The model usage is billed by Anthropic to you. We have no fees.

The audit runs on a fresh GitHub Actions runner in our public repository. It clones the exact commit, loads the corpus as a Claude Code plugin and walks it in one linear session: Read, Grep and read-only shell commands only, no subagents, nothing from the audited repository is ever built or executed. The run log is public.

The report is committed to our repository byte-for-byte and rendered here. The attester key writes one account per (repository, commit) on Solana devnet: repository hash, commit, corpus version, model, report sha256, severity counts and timestamp. The address derives from the URL and the commit, so no database is needed to find it.

Estimate, key, run, report, attestation. Nothing to install. The report becomes public (or private, by link) and is attested on Solana devnet.

Clone solanabr/auditor-skill, load it as a Claude Code plugin in your own repository and run the audit locally, on your subscription or key. No upload, no third party, nothing of ours involved.

git clone https://github.com/solanabr/auditor-skill
claude --plugin-dir ./auditor-skill
> /auditor:audit --scope program

Site, runner, attestation program and report pipeline in one repository, with the corpus as a pinned submodule. One command brings it up locally; the local runner audits any clone with your own Claude Code login.

git clone --recurse-submodules https://github.com/frcd10/auditor-attest.
cd auditor-attest && pnpm dev
pnpm audit:local -- Org/repo

Repository size, language mix, which checklists and known-vector groups apply, expected cost per model. No key needed.

Billed by Anthropic to your key at list prices. Measured on eight on-chain programs with Claude Opus 5: $16 to $32 each. Models offered: Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5.

Written by our attester key. No wallet or SOL on your side.

Attestations are what we intend to charge for later: a repository audited once would get a new attestation on every subsequent commit, backed by an audit of the diff since the last attested commit.