Capybara Nexus v2 is a frontier-class autonomous security auditing framework designed for the Gemini CLI. Built for professional Red-Teaming and Zero-Day Detection, it leverages advanced reasoning agents to discover complex, multi-stage vulnerabilities that traditional scanners miss. Unlike static analysis tools, Capybara Nexus v2 operates via a recursive cognitive loop and an empirical feedback mechanism, ensuring high-signal findings with near-zero false positives. This repository provides the elite-tier SKILL.md definitions and methodologies required to transform standard LLMs into expert-level security researchers capable of deep architectural analysis.
- Autonomous Exploit Chaining: Discovers how minor info leaks can be used to escalate attacks into full system compromises.
- Empirical Feedback Loop: Every finding is validated through a 3-cycle "Hypothesize -> Execute -> Revise" process.
- Structural Invariant Analysis: Identifies deep logic flaws in middleware, authentication, and tenant isolation.
- Glasswing Hardening: Provides architectural remediation strategies that eliminate entire vulnerability classes.
SKILL.md: The elite frontier-tier skill definition for Gemini CLI.METHODOLOGY.md: A detailed breakdown of the 5-stage Capybara reasoning chain.BENCHMARKS.md: Comparative analysis against standard LLM security prompting.DISCLAIMER.md: Legal and ethical usage guidelines.ATTACK_COVERAGE.md: MITRE ATT&CK and OWASP Top 10 mapping.nexus_audit.py: [NEW] The autonomous Python orchestrator for the Empirical Feedback Loop.gym/: [NEW] Multi-language test cases (C, Go, Python) with the Ablation Study.
- Installation: Copy
SKILL.mdto your Gemini CLI skills directory. - Activation: Run
activate_skill capybara-nexus-v2. - Autonomous Audit: Use the orchestrator to drive the reasoning loop:
python3 nexus_audit.py /path/to/target/project
- Proof of Power: Explore the
gym/directory to see why Capybara Nexus outperforms traditional scanners.
Inspired by the "Claude Mythos" model and Project Glasswing.