Pashov Solidity Security Skills

Professional-grade Solidity audit, fuzzing, and pre-audit recon skills for AI coding assistants.

Source repo
pashov/skills
Stars
★ 1.3k
Last updated
1mo ago
License
MIT
Primary language
JavaScript

At a glance

Works with
Portable with changesCodex · Claude CodeChatGPT (Partial support)
You'll need
Shell / CLINetwork accessLocal filesystem
Typical use
A smart contract developer who wants a quick self-check for high-severity issues before mainnet deployment.
Main limitation
The repo is prompt-and-skill content for other AI assistants, not a model or runtime; capability is capped by the platform you plug it into.
Maintenance
Commits in the last 6 monthslast commit 2026-10-21latest release v05102026 (2026-10-05)+2.0 ★/day over 7 days

What does this agent do, and when should you use it?

This is an open-source collection of AI skills from Pashov Audit Group focused on Solidity smart contract security. The repository ships three skill folders: solidity-auditor for AI-driven security audits, fizz to generate a complete Echidna/Medusa fuzz suite for Foundry or Hardhat projects, and x-ray for a pre-audit scan that produces a threat model, invariants, entry points, and git analysis. You use it by installing the repo into a skills-capable AI coding platform (Claude Code, Cursor, Codex, GitHub Copilot, Windsurf) and triggering a skill with a natural-language prompt. It is not a hosted audit service: everything runs inside your local development setup and relies on the model of whichever AI platform you use. The project is released under the MIT license.

After installation, prompts such as "run an x-ray on the codebase", "run the solidity auditor with all the different agents possible on *specified files*", or "run fizz on the codebase" dispatch to the matching skill. x-ray scans the codebase before a formal audit, producing a threat model, invariants, entry points, and git history analysis. solidity-auditor runs a Solidity security audit using multiple agents, which the README credits with finding hundreds of Critical/High vulnerabilities. fizz reads a Foundry or Hardhat project and generates a ready-to-run Echidna/Medusa fuzz suite. Skills are refreshed with the prompt "update skills to latest version". All analysis happens inside the AI platform and results come back as conversation output; the repository ships no standalone CLI or API.

  1. A smart contract developer who wants a quick self-check for high-severity issues before mainnet deployment.
  2. A security researcher onboarding to a new protocol who needs a threat model and invariant list fast.
  3. An audit team running x-ray as pre-audit recon to decide where manual review should focus.
  4. A Foundry or Hardhat project that needs Echidna/Medusa fuzz tests generated automatically.
  5. A Web3 team already using Claude Code or Cursor that wants security review folded into the daily coding loop.

How do you install or deploy this agent?

Hand the repository URL to an AI coding platform that supports installing skills or plugins. The README's install prompt is:

Install https://github.com/pashov/skills/ and run an x-ray on the codebase

Alternatively, clone the repo and mount the skill folders the way your platform expects:

git clone https://github.com/pashov/skills.git

The README does not document per-platform install commands, target directories, or required credentials; that information is missing.

How do you use this agent?

Once installed, trigger skills from inside your project directory with natural-language prompts:

run an x-ray on the codebase
run the solidity auditor with all the different agents possible on *specified files*
run fizz on the codebase

To pull the latest version:

update skills to latest version

You need a working account and model access on a supported AI platform.

What are this agent's strengths and limitations?

Pros
  • Three clearly separated skills form a pipeline: pre-audit recon (x-ray), multi-agent auditing (solidity-auditor), and fuzz suite generation (fizz).
  • Explicitly lists five supported platforms — Claude Code, Cursor, Codex, GitHub Copilot, and Windsurf.
  • fizz targets Foundry or Hardhat directly and emits Echidna/Medusa suites, fitting existing Solidity toolchains.
  • Maintained by a firm that specializes in smart contract security, under MIT so you can fork and adapt it.
Limitations
  • The repo is prompt-and-skill content for other AI assistants, not a model or runtime; capability is capped by the platform you plug it into.
  • The README gives no per-platform install commands or directory conventions, so setup requires trial and error.
  • The claim that solidity-auditor finds "hundreds of Critical/High vulnerabilities" has no published benchmark backing it.
  • No standalone CLI, API, or hosted service, so it cannot be wired into a CI pipeline for automated audits.

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

Agent Source review Form / cost Stars Updated Language Full support on
Pashov Solidity Security Skills This agent 45 · Major gaps — ★ 1.3k 1mo ago JavaScript Codex · Claude Code
Cloudflare Security Audit Skill 84 · Good Agent plugin / skillFree + model costs ★ 76k 2mo ago JavaScript —
RAMPART 78 · Good Library / SDKFree + model costs ★ 427 1d ago Python —
OrchestKit AI Development Toolkit 70 · Some gaps Agent plugin / skillFree + model costs ★ 305 today TypeScript Claude Code

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Major gaps
45/ 100 5-point scale 2.3 / 5
Trust 10/29
Reliability 8/14
Adaptability 10/18
Convention 8/18
Effectiveness 6/13
Verifiability 3/8
Why each dimension lost points
Trust10 / 29 · 1.7/5

Skills scan code read-only, but x-ray performs curl to fetch remote VERSION, runs forge/npx coverage, and writes to x-ray/ and .solidity-auditor/ — external side effects without explicit authorization scoping, so least_privilege and external_effects score 1. fizz offers guided/automatic modes and a cost-confirmation checkpoint, giving real user confirmation (2). Data-flow transparency is scattered with no unified statement of where data goes (1). Sensitive-data handling is a single CLAUDE.md line with no mechanism (1). Dependency security is entirely unaddressed — forge, npx, python3, Echidna/Medusa have no version pinning or verification (0). Rollback appears only in fizz-convert build-failure handling, narrow scope (1). Source attribution has README branding and MIT copyright line, but publisher identity is unverified (1).

Reliability8 / 14 · 2.9/5

SKILL.md internal rules are self-consistent (mode resolution, hard rules, action matrix), scoring 2. Dependency availability only says to report missing tooling, with no version constraints or install checks (1). Failure messaging is well designed (stop on step failure and report blocker, state what was attempted when tooling missing), scoring 2.

Adaptability10 / 18 · 2.8/5

Targets Solidity auditing on Foundry/Hardhat with clear audience and scenarios (2). Capability boundaries are defined via exclude patterns, parameters, and mode switching (2). Trigger words are specific in frontmatter (2). Environment fit covers only Foundry/Hardhat and specific AI platforms with no adaptation notes for others (1).

Convention8 / 18 · 2.2/5

Information architecture is clear (README table, per-skill directories, CLAUDE.md structure), scoring 2. Install notes are natural-language prompts only, with no version pinning or verification steps (1). Naming stability has VERSION and an auto-bump workflow but only for solidity-auditor; other skills unversioned (1). Examples/FAQ are only invocation prompts, no output samples or FAQ (1). Known limitations are barely declared (1). License is full MIT text (2). Versioning/changelog is only a self-incrementing VERSION number with no changelog (1). Maintenance responsibility points to a GitHub profile and community channels, but publisher is unverified (1).

Effectiveness6 / 13 · 2.3/5

Output usability is decent (report files, PROPERTIES.md, x-ray directory structure are explicit), scoring 2. Marginal value depends on the project; README's 'hundreds of Critical/High vulnerabilities' claim is unsupported (1). Cost-benefit has fizz cost confirmation and model tiering, but overall token/subagent overhead is large and unquantified (1).

Verifiability3 / 8 · 1.9/5

Claim traceability is weak: README vulnerability-count claims have no citation (1). Cross-source corroboration is limited; skills reference each other consistently but no external validation (1). Fact/inference separation is partially present (e.g., 'expected failure' notes) but claims and facts are mixed overall (1).

Risks and how to mitigate them
  • Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
  • The x-ray skill performs curl to fetch a remote VERSION and runs forge/npx coverage commands — external network and execution side effects; confirm network and toolchain permissions before use.
  • Dependencies (forge, npx, python3, Echidna/Medusa) have no version pinning or integrity checks, so behavior may vary across environments.
  • README's claim of 'hundreds of Critical/High vulnerabilities found' is unsupported and should not be treated as a quality guarantee.
  • Publisher identity is not verified by FollowAgents; maintenance responsibility and update path point only to a personal GitHub profile.
  • Versioning covers only solidity-auditor; other skills have no version number or changelog.
Evidence confidence: Low Reviewed Nov 15, 2026 Reviewed revision eacfc09b365d
See the full review method →
Maintain this project? Add the FARS badge to your README

The badge shows the current review score and verdict and links back to this page. It refreshes about once an hour after a re-review.

FARS

[![FARS review](https://followagents.com/badge/pashov-skills.svg)](https://followagents.com/en/agents/pashov-skills)
<a href="https://followagents.com/en/agents/pashov-skills"><img src="https://followagents.com/badge/pashov-skills.svg" alt="FARS review"></a>

Recent releases

Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.

  1. v05102026
    Release notes excerpt
    What's Changed * Update Fizz demo GIF by @0xfirefistt in https://github.com/pashov/skills/pull/38 * fizz: remove shipped self-test and de-hardcode eval paths (#33) by @0xfirefistt in https://github.com/pashov/skills/pull/39 * solidity-auditor: v4 — loop mode, scan memory, shell-assembled report by @0xfirefistt in https://github.com/pashov/skills/pull/48 * solidity-auditor: keep mental tools internal, drop mandatory marker protocol by @0xfirefistt in https://github.com/pashov/skills/pull/49 * solidity-auditor: bump VERSION to 4.1 by @0xfirefistt in https://github.com/pashov/skills/pull/52 …
    View on GitHub ↗
  2. v23092026
  3. v28062026

FAQ

Does using these skills cost money?
The repository is free under MIT, but you need an account and model access on a supported AI platform, and that platform bills you.
Can it replace a human security audit?
The README frames it as an audit aid (recon, bug discovery, fuzz test generation) and does not claim to replace manual review; a professional should still verify findings.
Which tech stacks does it cover?
Primarily Solidity smart contracts; fizz explicitly supports Foundry and Hardhat projects. Other languages are outside the documented scope.
Does my code leave my machine?
Skills run inside the chosen AI platform, so code is processed by that platform's models; the data flow depends on the platform, not on this repository.
Is Pashov Solidity Security Skills open source?
Yes. The source code is on GitHub (pashov/skills) under the MIT licence.
Is Pashov Solidity Security Skills still maintained?
Development has slowed: the latest commit on GitHub is from 2026-10-21, within the last six months but not the last 30 days. The latest release is v05102026 (2026-10-05).
Who makes Pashov Solidity Security Skills?
Pashov Solidity Security Skills is published on GitHub by the account pashov. FollowAgents has not verified which person or company is behind that account.
Was this page helpful?
View on GitHub ↗ Install ↓

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related concepts