agentleFS
mtarcuremtarcure/claude-vibe-squad/.agents/skills/evm-audit-flow/SKILL.md

evm-audit-flow

Skill220 starsChanged 19 days agoMITView on GitHub

Use when you have a Foundry or Hardhat repository of Solidity or Vyper contracts and want the end-to-end EVM review workflow on native CLIs — a slither static pass, semgrep taint rules, mythril symbolic execution, echidna and medusa invariant fuzzing, halmos formal checks, then a forge build-and-test regression.

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 2 are distinctive, and 9 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 9

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Start from pre-audit-threat-model if an x-ray does not already exist.

    line 13
  2. Very specific:

    Static analysis pass: Run slither . directly from the project root (it needs the Foundry/Hardhat build context, not a loose .sol file).

    line 17
  3. Very specific:

    Symbolic execution: Run myth analyze <contract> --execution-timeout 600 on every contract flagged HIGH by step 1, plus any contract with …

    line 43
  4. Very specific:

    Formal verification (critical paths only): Run halmos --function <prop> --loop <N> on high-value properties Echidna found hard to falsify …

    line 55
  5. Very specific:

    Echidna coverage plateau: add a corpus dir (--corpus-dir) or switch to Medusa's coverage-from-corpus mode.

    line 71
  6. Very specific:

    Do NOT report Slither medium/low findings without confirming exploitability via a call sequence or myth evidence.

    line 78
  7. Distinctive:

    Do NOT run Echidna/Medusa without invariant tests — property-less fuzzing yields nothing.

    line 80
  8. Very specific:

    Do NOT run myth on a proxy without resolving the implementation first.

    line 81
  9. Distinctive:

    Do NOT skip Slither for "simple" contracts — over half of historical EVM findings are in contracts under 200 lines.

    line 82
  10. Very specific:

    Do NOT invoke any dead tool_wrappers/* name (slither_scan, mythril_analyze, echidna_fuzz, halmos_check, forge_test).

    line 84
  11. Very specific:

    Do NOT route through the retired guarded-slither / guarded-semgrep MCPs (Trail of Bits mcp-context-protector wrappers, removed from both …

    line 87

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

evm-audit-flow

Stateful EVM audit pipeline from raw contract source through confirmed-finding delivery. Use when you have a Foundry or Hardhat project with Solidity/Vyper contracts. Every step runs a native host CLI (slither, semgrep, myth, echidna/medusa, halmos, forge) invoked directly; there is no MCP wrapper in front of them, no tool_wrappers layer, and no retired KG. Start from pre-audit-threat-model if an x-ray does not already exist.

Order of ops

  1. Static analysis pass. Run slither . directly from the project root (it needs the Foundry/Hardhat build context, not a loose .sol file). slither . --json - gives structured output, slither --list-detectors enumerates detectors, and slither . --print function-summary gives per-function context. Classify detectors: HIGH-signal (reentrancy-eth, arbitrary-send, controlled-delegatecall, suicidal, unprotected-upgrade) vs NOISE (naming-convention, too-many-digits, solc-version). Do NOT dismiss reentrancy-eth as a false positive without tracing the call stack. solc-select is available for pinning the compiler, but do not pass --always-install (no silent network downloads); confirm the version is already present.
  2. Pattern/taint pass (the static analog of symbolic execution). Run project-specific Semgrep rules with the semgrep CLI directly — semgrep scan --config <rules.yaml> --metrics=off --json <path> — taint sources = external/msg.sender-reachable inputs, sinks = value transfers, delegatecall, storage writes. Author rules with semgrep-rule-author; a reachable sink with no realized impact is a lead, not a finding. Local rule files need no login.

    A zero is not a clean scan until the same rule has fired on a known positive. Two measured ways to get an empty result (2026-10-27, semgrep 1.157.0): a rule file that fails YAML parsing exits 7 and reports the error only in the JSON errors array with an empty results list (an unquoted pattern containing { or : is the usual cause — quote it: pattern: "$X.call{value: $V}($A)" fires, the unquoted form does not parse); and a pattern that does not match the call shape — $X.call(...) misses to.call{value: 1}("") — returns zero findings, zero errors, exit 0. Check the exit code, errors, and the positive control before believing a zero.

  3. Symbolic execution. Run myth analyze <contract> --execution-timeout 600 on every contract flagged HIGH by step 1, plus any contract with external calls, delegatecall, or selfdestruct. Surface: integer overflow, unchecked return values, tx.origin access-control bypass, ether-locking. For proxies, resolve the implementation address first — myth cannot follow delegatecall to a separate contract by default.
  4. Invariant authoring + fuzzing. Author/refresh Echidna or Medusa invariants (token conservation, access-control, CEI reentrancy) in test/invariants/. Run echidna . --contract <Test> --config echidna.yaml for an initial pass; if coverage plateaus switch to medusa fuzz --config medusa.json (coverage-guided). For DeFi, apply defi-invariant-check. Minimum meaningful campaign: 30 minutes.
  5. Formal verification (critical paths only). Run halmos --function <prop> --loop <N> on high-value properties Echidna found hard to falsify in bounded time. Halmos gives unbounded symbolic verification for simpler properties; do not throw the whole suite at it (it times out on complex state machines or unbounded loops).
  6. Lint + compile regression. forge build (or solc on modified contracts) to confirm no compilation regression; forge fmt --check for style.
  7. Final test run. forge test to confirm the suite passes; record any new failure.

When to pivot

  • Slither false-positive flood: narrow with --detect <high-signal list> and add a .slither.config.json suppression for known project FPs.
  • myth times out on complex contracts: --execution-timeout 300 --create-timeout 60, and focus on functions with external calls and value transfers.
  • Echidna coverage plateau: add a corpus dir (--corpus-dir) or switch to Medusa's coverage-from-corpus mode.
  • Halmos timeouts on loop-heavy code: bound with --loop; Halmos is not for unbounded loops.

Anti-patterns

  • Do NOT report Slither medium/low findings without confirming exploitability via a call sequence or myth evidence.
  • Do NOT run Echidna/Medusa without invariant tests — property-less fuzzing yields nothing.
  • Do NOT run myth on a proxy without resolving the implementation first.
  • Do NOT skip Slither for "simple" contracts — over half of historical EVM findings are in contracts under 200 lines.
  • Do NOT invoke any dead tool_wrappers/* name (slither_scan, mythril_analyze, echidna_fuzz, halmos_check, forge_test); those wrap a container library that no longer exists — use the native CLIs above.
  • Do NOT route through the retired guarded-slither / guarded-semgrep MCPs (Trail of Bits mcp-context-protector wrappers, removed from both lanes on 2026-10-27, commit 02442bdc); their tool names (run_detectors, semgrep_scan_with_custom_rule) no longer resolve — run the CLIs above.

Example

# Step 1: static analysis (native)
slither . --exclude naming-convention,solc-version

# Step 2: taint rules (native) -- run the same rule on a known positive first
semgrep scan --config rules/evm-taint.yaml --metrics=off --json src/

# Step 3: symbolic execution on a flagged contract
myth analyze src/Vault.sol --execution-timeout 600

# Step 4: fuzzing (30-min minimum for DeFi)
echidna . --contract InvariantTest --config echidna.yaml --test-limit 500000

# Step 7: final test run
forge test

Recording (chrono-vault)

The task packet's injected memory contract owns the exact call shape, sequence, and fields - see wirework-reflect. Do not copy a record(...) example or add fields (including source_task) from memory; the server binds them, and a baked example violates the run's authenticated schema. Memory is best-effort telemetry and never gates the work. What is worth recording here is the task-specific outcome: finding counts by severity, invariant violations, tools run.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.