Zeus Auditor

Zeus is an AI-first security auditor specifically designed for GenLayer Intelligent Contracts. It runs a structured 4-phase audit pipeline to detect vulnerabilities unique to LLM integration, web connectivity, the Equivalence Principle, and the GenVM execution environment.

Executable discipline for AI coding agents - skills the agent loads on demand and applies while it works, not prose guides it ignores.

Install

npx skills - recommended for any Agent Skills host

Works with any compatible agent (Claude Code, Cursor, Copilot, Gemini CLI, OpenCode, Codex, and more). Each command is independent - run the one(s) you want:

npx skills add https://github.com/Wizbisy/zeus

Claude Code

/plugin marketplace add Wizbisy/zeus
/plugin install zeus@Wizbisy

Codex

codex plugin marketplace add Wizbisy/zeus

Then run codex, open /plugins, select zeus, and install it.

For other hosts, expand below.

git clone https://github.com/Wizbisy/zeus.git
ln -s "$(pwd)/zeus/skills/zeus" ~/.cursor/skills/zeus

git clone https://github.com/Wizbisy/zeus.git ~/.copilot/skills/zeus

git clone https://github.com/Wizbisy/zeus.git ~/.agents/skills/zeus

git clone https://github.com/Wizbisy/zeus.git
ln -s "$(pwd)/zeus/skills/zeus" ~/.gemini/config/skills/zeus

https://github.com/Wizbisy/zeus

What changes with the plugin

Routes GenLayer audits to a rigorous 4-phase methodology instead of blind text-based guessing. Ensures GenVM and Equivalence Principle specifics are respected.

tldr - what changes with the plugin:

Try:

• /zeus
• /zeus-quick
• /zeus-fuzz
• /zeus-review
• /zeus-init

Features & Detection Lenses

• GenLayer Native: Specifically built for Python contracts using the genlayer SDK (gl.Contract, gl.nondet.*, gl.eq_principle.*).
• 4-Lens Detection: Analyzes contracts for:
  • Lens A (AI-01→03): Prompt Injection & LLM Output Safety
  • Lens B (WEB-01→02): Web Data & SSRF
  • Lens C (EQ-01→03): Equivalence Principle & Consensus Logic
  • Lens D (ST-01→03, MSG-01): State, Storage & Access Control
• State Auditor: Tracks state mutations driven by non-deterministic outputs and verifies appropriate GenVM storage types (TreeMap, DynArray, u256).
• Strict Verification (Critic): 6 kill gates automatically destroy false positives. Gate 2 requires demonstrable sanitization — response_format="json" alone does not kill injection findings. Gate 6 exempts AI-* and WEB-* categories from low-impact filtering.
• Mandatory Reviewer: Every audit runs a devil's advocate phase on killed findings to catch over-filtering before the final report.
• Property-Based Fuzzing: Auto-generates Python test scripts that mock non-deterministic LLM/web APIs with adversarial payloads to fuzz contract invariants.
• Test Fixtures: Runnable .py contracts with .expected.md golden reports for reproducible audit verification (see tests/fixtures/).

Why this plugin

• Honest by construction. Any findings killed by the 6 kill gates are reviewed by a mandatory devil's advocate phase. You can also re-run the review via /zeus-review.
• Reproducible. Runnable test fixtures with golden expected reports let you verify the pipeline produces correct results.
• Token-lean. A short SKILL.md routes to reference files that load only when the task needs them. The agent does not carry the whole guide in context.
• Portable. One discipline across Claude Code, Cursor, Copilot, Antigravity, OpenCode, Codex, and Kiro.

Authors & Attribution

Zeus is built by Wizbisy.

The pipeline architecture and verification methodology were adapted from Krait (by Zealynx Security). For full details, see ATTRIBUTION.md.

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.