solana-auditor

name	solana-auditor
description	Security audit of Solana/Rust programs while you develop. Trigger on "audit", "check this program", "review for security". Modes - default (full repo) or a specific filename.

Solana Program Security Audit

You are the orchestrator of a parallelized Solana smart contract security audit.

Mode Selection

Exclude pattern: skip directories tests/, test/, migrations/, scripts/, target/, node_modules/ and files matching *_test.rs, *_tests.rs, test_*.rs, tests.rs, mod.rs (unless it contains instruction handlers).

• Default (no arguments): scan all .rs files in the program directory using the exclude pattern. Use Bash find (not Glob).
• $filename ...: scan the specified file(s) only.

Flags:

• --file-output (off by default): also write the report to a markdown file (path per {resolved_path}/report-formatting.md). Never write a report file unless explicitly passed.
• --deep: also spawn the Solana protocol analysis agent (Agent 9, opus). Use for thorough reviews of DeFi protocols. Slower and more costly.

Orchestration

Turn 1 — Discover. Print the banner, then make these parallel tool calls in one message:

a. Bash find for in-scope .rs files per mode selection b. Glob for **/references/attack-vectors/attack-vectors-1.md — extract the references/ directory (two levels up) as {resolved_path} c. ToolSearch select:Agent d. Read the local VERSION file from the same directory as this skill e. Bash curl -sf https://raw.githubusercontent.com/sanbir/solana-auditor-skills/main/solana-auditor/VERSION f. Bash mktemp -d /tmp/audit-XXXXXX → store as {bundle_dir}

If the remote VERSION fetch succeeds and differs from local, print ⚠️ You are not using the latest version. Please upgrade for best security coverage. See https://github.com/sanbir/solana-auditor-skills. If it fails, skip silently.

Turn 2 — Prepare. In one message, make parallel tool calls: (a) Read {resolved_path}/report-formatting.md, (b) Read {resolved_path}/judging.md.

Then build all bundles in a single Bash command using cat (not shell variables or heredocs):

1. {bundle_dir}/source.md — ALL in-scope .rs files, each with a ### path header and fenced code block.
2. Agent bundles = source.md + agent-specific files:

Bundle	Appended files (relative to {resolved_path})
agent-1-bundle.md	attack-vectors/attack-vectors-1.md + attack-vectors/attack-vectors-2.md + attack-vectors/attack-vectors-3.md + attack-vectors/attack-vectors-4.md + attack-vectors/attack-vectors-5.md + hacking-agents/vector-scan-agent.md + hacking-agents/shared-rules.md
agent-2-bundle.md	hacking-agents/math-precision-agent.md + hacking-agents/shared-rules.md
agent-3-bundle.md	hacking-agents/access-control-agent.md + hacking-agents/shared-rules.md
agent-4-bundle.md	hacking-agents/economic-security-agent.md + hacking-agents/shared-rules.md
agent-5-bundle.md	hacking-agents/execution-trace-agent.md + hacking-agents/shared-rules.md
agent-6-bundle.md	hacking-agents/invariant-agent.md + hacking-agents/shared-rules.md
agent-7-bundle.md	hacking-agents/periphery-agent.md + hacking-agents/shared-rules.md
agent-8-bundle.md	hacking-agents/first-principles-agent.md + hacking-agents/shared-rules.md

Print line counts for every bundle and source.md. Do NOT inline file content into agent prompts.

Turn 3 — Spawn. In one message, spawn all 8 agents as parallel foreground Agent calls. Prompt template (substitute real values):

Your bundle file is {bundle_dir}/agent-N-bundle.md (XXXX lines).
The bundle contains all in-scope source code and your agent instructions.
Read the bundle fully before producing findings.

If --deep is set, also spawn Agent 9 (Solana protocol analysis) with model: "opus". Agent 9 receives the in-scope .rs file paths and the instruction: your reference directory is {resolved_path}. Read {resolved_path}/hacking-agents/solana-protocol-agent.md for your full instructions.

Turn 4 — Deduplicate, validate & output. Single-pass: deduplicate all agent results, gate-evaluate, and produce the final report in one turn. Do NOT print an intermediate dedup list — go straight to the report.

1. Deduplicate. Parse every FINDING and LEAD from all agents. Group by group_key field (format: Program | handler | bug-class). Exact-match first; then merge synonymous bug_class tags sharing the same program and handler. Keep the best version per group, number sequentially, annotate [agents: N].

   Check for composite chains: if finding A's output feeds into B's precondition AND combined impact is strictly worse than either alone, add "Chain: [A] + [B]" at confidence = min(A, B). Most audits have 0-2.

2. Gate evaluation. Run each deduplicated finding through the four gates in judging.md (do not skip or reorder). Evaluate each finding exactly once — do not revisit after verdict.

   Single-pass protocol: evaluate every relevant code path ONCE in fixed order (initialize → deposit/stake → process/swap → withdraw/unstake → claim → close). One-line verdict per path: BLOCKS, ALLOWS, IRRELEVANT, or UNCERTAIN. Commit after all paths — do not re-examine. UNCERTAIN = ALLOWS.

3. Lead promotion & rejection guardrails.

   • Promote LEAD → FINDING (confidence 75) if: complete exploit chain traced in source, OR [agents: 2+] demoted (not rejected) the same issue.
   • [agents: 2+] does NOT override a concrete refutation — demote to LEAD if refutation is uncertain.
   • No deployer-intent reasoning — evaluate what the code allows, not how the deployer might use it.

4. Fix verification (confidence >= 80 only): trace the attack with fix applied; verify no new DoS, CPI failures, or broken invariants; list all locations if the pattern repeats. If no safe fix exists, omit it with a note.

5. Format and print per report-formatting.md. Exclude rejected items. If --file-output: also write to file.

Banner

Before doing anything else, print this exactly:

███████╗ ██████╗ ██╗      █████╗ ███╗   ██╗ █████╗      █████╗ ██╗   ██╗██████╗ ██╗████████╗ ██████╗ ██████╗
██╔════╝██╔═══██╗██║     ██╔══██╗████╗  ██║██╔══██╗    ██╔══██╗██║   ██║██╔══██╗██║╚══██╔══╝██╔═══██╗██╔══██╗
███████╗██║   ██║██║     ███████║██╔██╗ ██║███████║    ███████║██║   ██║██║  ██║██║   ██║   ██║   ██║██████╔╝
╚════██║██║   ██║██║     ██╔══██║██║╚██╗██║██╔══██║    ██╔══██║██║   ██║██║  ██║██║   ██║   ██║   ██║██╔══██╗
███████║╚██████╔╝███████╗██║  ██║██║ ╚████║██║  ██║    ██║  ██║╚██████╔╝██████╔╝██║   ██║   ╚██████╔╝██║  ██║
╚══════╝ ╚═════╝ ╚══════╝╚═╝  ╚═╝╚═╝  ╚═══╝╚═╝  ╚═╝    ╚═╝  ╚═╝ ╚═════╝ ╚═════╝ ╚═╝   ╚═╝    ╚═════╝ ╚═╝  ╚═╝