Reference copy of the operating contract for the defense lens at a human gate — input binding, the six sources of the case for crossing, the untrusted-input and exhibit-marker rules, the evidence whitelist, the degraded-input table, the severity/confidence/variance scales, the output contract, and the escalation table. NOT preloaded into any agent. agents/iron-loop/advocate-critic.md carries this contract in its own body; this file is loadable on demand through the Skill tool by any agent that needs to read the same rules.
Reviews AI-generated code for common pitfalls — over-engineering, missing edge cases, fabricated patterns, hallucinated imports, stale framework idioms, vacuous tests.
Detects AI-generated code that references non-existent packages, APIs, methods, or fabricated patterns.
Paranoid LLM red-team analyst — scans applications that call LLMs for OWASP LLM Top 10 v2 (2025) findings and maps them to MITRE ATLAS v5.4.0 adversary tactics.
Builds the import graph and detects circular dependencies, layer violations, instability mismatches, and cross-module coupling.
Detects which architecture pattern a codebase follows (Layered, Hexagonal, Clean, Onion, CQRS, DDD, Vertical Slice, Modular Monolith, Microservices) with confidence scoring across 7 languages.
AI regulatory governance scanner — classifies AI systems against EU AI Act risk tiers, NIST AI RMF / AI 600-1 functions, and ISO/IEC 42001 controls; flags missing inventory, oversight, documentation, transparency, and conformity-assessment artifacts.
Validates audit logging for compliance, security, and operational visibility.