← all repositories
jmanico/grim-prompter

Your AI agent can now do the threat modeling you kept postponing

Grim Prompter turns Claude Code or Codex into a disciplined threat-modeling analyst that cites file and line for every claim.

grim-prompter
Too new to rank. Velocity needs a few days of history.
not enough history yet
star history

What it does It’s a library of prompts and a runbook that your existing coding agent loads as a skill. You point it at a repo, pick an architecture (36 of them, from web apps to OT/ICS to RAG pipelines), and the agent walks a structured review: scope, trust boundaries, threats, existing controls, and mitigations. Output lands in a threat-model/ folder with a human-readable report plus JSON artifacts you can use as a baseline for future change reviews.

The interesting bit The discipline is the product. The runbook demands read-only inspection and file-and-line citations, and the output explicitly records what was skipped and which trust boundaries went unmapped — so you can audit the model instead of trusting it. There’s also a privacy track (LINDDUN GO/PRO, GDPR/CCPA profiles) that looks for harms to people even when the processing is technically authorized.

Key highlights

  • 36 architecture workflows and 29 method/analysis prompts, including STRIDE, LINDDUN, PASTA-style deep runs
  • Six-stage deep workflow where each stage reads the previous stage’s file — resumable and traceable
  • Change reviews: diff a branch against a saved JSON baseline instead of re-modeling from scratch
  • MCP-server reviews map tools to backend credentials and flag shared-token blast radius
  • Works with any agent via an AGENTS.md snippet, not just Claude Code and Codex

Caveats

  • The prompts are explicitly drafts, and the README is blunt that a threat model does not certify a system is secure — you’re expected to challenge every finding
  • Reports can contain sensitive details; the bundled .gitignore only covers runs inside the checkout, so you’re on your own elsewhere
  • The agent won’t invent risk ratings or approvals; you supply risk criteria and priorities yourself

Verdict Worth a look for any team already living in Claude Code or Codex that treats threat modeling as a chore. If you want a tool that outputs a rubber stamp, this is deliberately not that.

Frequently asked

What is jmanico/grim-prompter?
Grim Prompter turns Claude Code or Codex into a disciplined threat-modeling analyst that cites file and line for every claim.
Is grim-prompter open source?
Yes — jmanico/grim-prompter is open source, released under the MIT license.
How popular is grim-prompter?
jmanico/grim-prompter has 640 stars on GitHub.
Where can I find grim-prompter?
jmanico/grim-prompter is on GitHub at https://github.com/jmanico/grim-prompter.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.