Solana Vulnerability Scanner checks programs for arbitrary CPI, improper PDA validation, missing ownership checks, missing signer checks, sysvar account spoofing and improper instruction introspection.
Each report opens with a coverage table giving every pattern a verdict of found, clear or n/a with evidence, followed by findings that include the vulnerable code, an attack scenario and a fix in native or Anchor style.
What You Get
- A coverage table with all six patterns and a verdict with evidence for each
- Findings with severity, location, vulnerable code, attack scenario and recommendation
- Checklists for CPI, PDA, account validation, signer validation, sysvar and introspection
- Unit and Anchor integration test examples for malicious inputs
Common Use Cases
- Auditing a Solana vault or token program before launch
- Reviewing cross-program invocation logic for unvalidated program accounts
- Checking PDA derivation and canonical bump use
- Verifying Anchor account constraints for owner and signer checks
What Makes It Unique
The skill keys on Solana’s account model and requires a verdict for all six patterns, including a version-scoped sysvar row that must cite the targeted solana-program version. It rejects shortcuts such as assuming Anchor handles validation, and asks for the constraint or attribute by name, since an UncheckedAccount opts out of checks and a mut attribute is no ownership check.
Under The Hood
The skill is an instruction file with a linked pattern resource. The agent searches the repository with ripgrep for invoke, invoke_signed, PDA functions, deserialization and introspection calls, reads the matching code and may suggest Cargo lint settings and local validator tests with solana-test-validator and anchor test. It reads project files and writes its report as a response. It does not fetch web pages or call remote services.