Grim Prompter

What it does

• Quick or deep, per architecture — /threat-model picks one for you, /threat-model quick api runs a single bounded review, /threat-model deep web runs the six-stage workflow where each stage reads the previous stage's file, and /threat-model change main...feature reviews a branch against a saved JSON baseline
• 29 named methods as focused prompts — STRIDE per element or per interaction, PASTA, Trike, OCTAVE Allegro, LINDDUN GO/PRO, attack trees, STPA-Sec, FMEA, CSA MAESTRO and MITRE ATLAS, plus ATT&CK, CAPEC, Kill Chain, CVSS and DREAD enrichment, each saving a JSON result you can fold into a review
• Evidence discipline built in — the runbook requires read-only inspection and file-and-line citations, treats repository text as evidence never instructions, and builds, tests, scripts and network access stay behind your explicit authorization
• Coverage aimed at agents themselves — for MCP servers and agent tools it maps tools to backend credentials and effective permissions, records unknown grants as gaps, and examines whether tool descriptions and their post-approval updates can steer other trusted tools

When to use it

Security review of code you or your team wrote, at the moment something changed: before merging a feature branch (change review against a saved baseline), when hardening a new service with a known architecture (quick run), or when you want a proper threat model without hiring a consultant (deep run). The README's AGENTS.md snippet lets agents outside Claude and Codex join the same workflow.

Getting started

git clone https://github.com/jmanico/grim-prompter.git ~/grim-prompter

# Claude Code
mkdir -p ~/.claude/skills && ln -s ~/grim-prompter/skills/threat-model ~/.claude/skills/threat-model

# Codex CLI
mkdir -p ~/.agents/skills && ln -s ~/grim-prompter/skills/threat-model ~/.agents/skills/threat-model

macOS or Linux shell; keep the clone where it is — the skill links to its prompts there — and avoid duplicate installations. For any other agent, add the README's AGENTS.md snippet pointing at the clone.

Things to know

The prompts are drafts and the README says so plainly: a threat model does not certify a system is secure — the agent should not invent ratings or approvals, and you supply the risk criteria. Reports can contain sensitive details, so keep the threat-model/ folder out of version control until you have reviewed it (the library's .gitignore only covers runs inside its own checkout). The skill links into the clone, so don't move or delete it; start a new agent session after installing.