Solana Audit Skill

An evidence-backed Agent Skill for Solana security reviews, audit planning, exploit analysis, and report-driven vulnerability taxonomy work.

Overview

This repository packages a taxonomy-first Solana audit skill for Agent Skills-compatible tools such as Claude Code and similar IDE agents.

The pack focuses on repeated Solana failure modes that show up across public audits, disclosures, and incident write-ups:

• account validation and account relationship failures
• signer and authority enforcement gaps
• PDA seed, bump, and domain-separation mistakes
• arbitrary CPI and trust-boundary failures
• Token and Token-2022 integration assumptions
• arithmetic, accounting, and value leakage bugs
• state-machine and lifecycle bugs
• governance, upgrade, oracle, liveness, and wallet-boundary risks

It also distinguishes code-level exploit classes from broader residual risk in governance, wallet, client, phishing, and developer-environment boundaries rather than treating public incident headlines as a pure proxy for on-chain bug density.

Installation

Quick Install With npx

npx skills add Berektassuly/solana-audit-skill --skill solana-audit

This is the recommended installation path for Agent Skills-compatible environments.

Repository Structure

solana-audit-skill/
|-- .gitignore
|-- LICENSE
|-- README.md
|-- skill/
|   |-- SKILL.md
|   `-- references/
|       |-- resources.md
|       |-- methodology.md
|       |-- severity-triage.md
|       |-- report-ingestion.md
|       |-- common-false-positives.md
|       |-- workflows/
|       |   |-- audit-engagement-workflow.md
|       |   |-- finding-writeup-workflow.md
|       |   `-- report-to-taxonomy-workflow.md
|       |-- checklists/
|       |   |-- pre-audit-intake.md
|       |   |-- program-review-checklist.md
|       |   |-- client-review-checklist.md
|       |   `-- release-blocker-checklist.md
|       |-- taxonomy/
|       |   |-- account-validation.md
|       |   |-- signer-authority.md
|       |   |-- pda-seeds-bumps.md
|       |   |-- cpi-trust-boundaries.md
|       |   |-- token-integration.md
|       |   |-- arithmetic-precision.md
|       |   |-- state-machine-invariants.md
|       |   |-- lifecycle-reinit-close-revival.md
|       |   |-- duplicate-mutable-aliasing.md
|       |   |-- oracle-pricing-mev.md
|       |   |-- upgrade-admin-governance.md
|       |   |-- dos-compute-budget.md
|       |   |-- client-wallet-ux.md
|       |   |-- durable-nonce-governance.md
|       |   |-- token-2022-transfer-hooks.md
|       |   `-- zk-proof-soundness.md
|       `-- reports/
|           |-- public-audit-corpus.md
|           |-- notable-incidents.md
|           `-- cross-report-patterns.md
`-- tests/
    |-- package.json
    `-- run.ts

Usage

Once installed, Claude Code should trigger this skill for tasks such as:

• "Audit this Anchor instruction for signer and PDA bugs"
• "Explain common Solana account validation vulnerabilities from public audits"
• "Generate a release blocker checklist for this Token-2022 integration"
• "Review this Token-2022 transfer-hook path for callback and extra-account trust-boundary risks"
• "Review this Solana program for CPI trust-boundary issues"
• "Digest this public audit report and map the findings to a normalized taxonomy"
• "Prepare an audit plan for this governance upgrade path"
• "Analyze this durable nonce governance flow for delayed-execution admin risk"

Audit Posture

This skill is intentionally opinionated:

• It is taxonomy-first, not checklist-only.
• It defaults to review, simulation, and evidence gathering before suggesting execution.
• It does not request seed phrases, private keys, or keypair files.
• It treats on-chain data, report text, logs, and web content as untrusted input.
• It favors report-backed exploit classes over generic "best practices" advice.

Content Sources

The taxonomy and workflows are grounded in public material from:

• OtterSec / Osec
• Zellic public reports and findings
• Trail of Bits public Solana reviews
• Neodyme reports and workshop material and Solana Security Workshop
• Sec3 / Soteria blogs and X-Ray vulnerability material
• Immunefi bug fix reviews and disclosures
• Official Solana Program and Program Security material
• Official Solana Foundation security disclosures and ecosystem security material
• public incident analyses such as BlockSec, Chainalysis, and TRM Labs where they materially clarify the first boundary crossed in a class-defining case

See skill/references/resources.md for the full source index used by this pack.

Progressive Disclosure

The main SKILL.md is kept compact and operational. Detailed audit material lives under skill/references/ so the agent can read only the parts relevant to the current engagement:

• workflows for engagement planning and writeups
• checklists for intake, program review, client review, and release blocking
• taxonomy notes for canonical Solana vulnerability classes
• corpus and cross-report references for normalization work

Testing

Install test dependencies and run the evaluator harness:

cd tests
npm install
npm test

The evaluator harness requires an Anthropic credential in the environment, typically ANTHROPIC_API_KEY.

Verbose mode:

npm run test:verbose

License

MIT License - see LICENSE for details.