Grim Prompter

Grim Prompter is an MIT-licensed prompt library by Jim Manico that lets AI coding agents such as Claude Code and Codex threat model a project. It provides 36 architecture workflows and 29 method and analysis prompts, and the agent follows a structured review of code, trust boundaries, threats, existing controls, and proposed mitigations with checkable evidence.

What It Is

Grim Prompter is a set of Markdown prompts, a runbook, and a threat-model skill that you link into your coding agent. You open your project in the agent and ask for a quick or deep threat model, a single method such as STRIDE per element, or a review of how a branch changes an existing model. For agents without skill support, a project snippet for AGENTS.md, CLAUDE.md, or GEMINI.md points the agent at the runbook.

How a Run Works

The runbook requires read-only inspection and file-and-line citations; builds, tests, scripts, installs, and network access need explicit authorization. Results are written to a new folder under threat-model/, containing SYSTEM_UNDER_REVIEW.md (scope, evidence, assumptions, gaps), report.md with findings, mitigations, and verification plans, and JSON artifacts for later reviews. Deep runs go through six stages: scope, system modeling, threat analysis, challenge, treatment planning, and reporting, with each stage reading the previous stage's file.

Architectures and Methods

Architecture prompts cover applications, application structure, cloud and platforms, data and AI, devices and physical systems, identity, money and collaboration, and distributed systems, including RAG and agentic AI. Method prompts include STRIDE, PASTA, Trike, LINDDUN, attack trees, STPA, FMEA, CSA MAESTRO, MITRE ATLAS, ATT&CK, CAPEC, CVSS, and DREAD. Helper prompts cover evidence gathering, architecture and method selection, model challenging, and change review. For MCP servers and agent tools, the review maps tools to credentials and permissions where evidence supports it and records unknowns as gaps.

Review Expectations

The README states the prompts are drafts, that a threat model does not certify a system as secure, and that risk ratings and approvals should come from the user rather than the agent. Reports should be reviewed for sensitive details before being committed to version control.