ntd25022006q/deerflow — reverse-engineered prompt

Reverse engineered prompt

GitHub

Build me a TypeScript Node.js tool called DeerFlow that helps keep AI coding agents and projects from cutting corners.

I want it to run from the command line and check a project through a strict 21 gate quality pipeline. It should catch things like unsafe file changes, bad patterns, mock data, secrets, dependency issues, weak tests, security problems, type errors, lint errors, missing evidence, and incomplete work. If a gate fails, the tool should clearly say what failed and lock the agent or workflow until it’s fixed.

Also include a simple workflow engine for agent tasks, a guard that monitors agent behavior, pluggable skill modules like code review, security, tests, UI, and search, plus a basic memory system that can remember useful conversation info using TF IDF style matching.

Use TypeScript, Node 20 or newer, tests, linting, formatting, and a CLI with commands for quality gate, compliance, and agent guard. Keep it production ready and include clear setup instructions.

Want more depth? Deep Reverse