Threat-Model: Instructions file for Claude Code

AGENTS.md — Threat Modeling Repo

This file is read automatically by AGENTS.md-compliant coding agents (OpenAI Codex CLI, and others that follow the same convention). It's the agent-agnostic entry point into the same workflow that's packaged as a Claude Code / Cowork plugin (.claude-plugin/) and as Cursor project rules (.cursor/rules/threat-modeling.mdc) elsewhere in this repo — all three point at the same skills/threat-modeling/ instructions and scripts, so behavior is consistent regardless of which agent is driving.

What this repo does

Runs an end-to-end security threat model for a service or system — conventional application/API infrastructure and AI/ML/LLM infrastructure alike — pulling live context from Jira, Jenkins, ServiceNow, and a code repository, then producing:

• A STRIDE threat register scored for likelihood and business impact (optionally the full 7-stage PASTA process for higher-rigor engagements)
• AI/ML-specific threats (prompt injection, training/RAG data poisoning, model extraction, excessive agency, insecure output handling, unbounded consumption) when the system has AI/ML components
• Every threat mapped to OWASP Top 10 (2021), OWASP API Security Top 10 (2023), and/or OWASP Top 10 for LLM Applications (2025)
• Attack-chain mapping with MITRE ATT&CK (Enterprise) and MITRE ATLAS technique tagging
• Compensating-controls, mitigation-assurance, and compliance-gap analysis
• A draw.io-compatible DFD, a draw.io-compatible threat-model diagram, a Word report, and/or an Excel risk register

Before doing anything: read the skill

Read skills/threat-modeling/SKILL.md in full and follow its 9 steps in order. It is the canonical, tool-agnostic instruction set — do not shortcut to writing a threat list without walking Steps 1-3 (scoping, context pull, DFD/trust-boundary modeling) first. Its references/ directory has the full methodology detail:

Reference	Covers
references/stride-methodology.md	The six STRIDE categories, minimum-threats rule, Likelihood x Impact scoring
references/pasta-methodology.md	Full 7-stage PASTA process for higher-rigor engagements
references/ai-threat-taxonomy.md	AI/ML/LLM pipeline threats layered onto STRIDE
references/owasp-mappings.md	OWASP Top 10 / API Security Top 10 / LLM Top 10 ID tables
references/mitre-mappings.md	MITRE ATT&CK and ATLAS tactic/technique tagging
references/attack-chain-mapping.md	How to build a valid multi-step attack chain
references/compensating-controls-and-gaps.md	Compensating controls, residual risk, compliance gaps
references/report-data-schema.json	The JSON shape the report-builder scripts consume
references/examples/	Worked examples: sample_* (conventional app), sample_ai_* (AI/RAG/agent system)