Anchor Pre-Audit Sweep

Audit findings cluster. A large share of what a Solana audit returns is the same handful of missing checks: an authority typed AccountInfo instead of Signer, a token account whose owner is never constrained, PDA seeds that scope to a constant instead of a user, arithmetic running unchecked because nobody set overflow-checks in the release profile.

Those are mechanical to look for and expensive to be told. This sweeps the whole programs tree for them and returns a table with severity, file:line, and — the part that decides whether anything gets fixed — what a caller actually gains from each one. "Missing signer check" is a category. "Any caller can set the fee recipient to their own account" is a finding.

It enumerates the full surface first and refuses to sample, because a missing check is invisible in the file you did not open. Two sections are mandatory in the output: what could not be reached, and what is out of scope by design. A sweep that quietly skipped the CPI review reads identical to one that found nothing there.

This does not replace an audit. It clears the floor so the audit spends its time on your design instead of your boilerplate. Ships with the check catalogue as CSV for CI, and a shell script that enumerates the surface.