A lightweight framework for AI-assisted software development. Drop it into any project and get structured, repeatable results from any LLM.
CLAUDE.md — The brain. A single file that governs how AI works on your project: rules, charter, roadmap, and session state. It survives context resets and keeps the AI on track across long sessions.
.agents/ — Specialized auditor prompts you can feed to any LLM:
| Agent | What it does |
|---|---|
audit-security |
Finds vulnerabilities — SQLi, XSS, CSRF, hardcoded secrets |
audit-archeologist |
Maps business logic, finds dead code, proposes clean architecture |
audit-database |
Schema issues, missing indexes, N+1 queries, data integrity |
audit-code-quality |
God objects, duplication, naming, error handling |
audit-frontend |
Responsive issues, accessibility, touch targets, UI consistency |
audit-ops |
Deployment, backups, SSL, monitoring, disaster recovery |
audit-prospector |
Finds existing patterns before you write new code |
KB/ — A modular knowledge base that grows with your project. Keeps context light by loading only what's needed.
- Copy the framework file into your project root
- Fill in the Charter section (project name, stack, goals)
- Copy
.agents/for auditor capabilities - Start a session — the AI reads the file and follows the rules
The framework works with any LLM CLI that reads a project-level instruction file. Just use the right filename:
| LLM | File | CLI |
|---|---|---|
| Claude | CLAUDE.md |
Claude Code |
| Gemini | GEMINI.md |
Gemini CLI |
| ChatGPT | CODEX.md |
Codex CLI |
| Other | Check your CLI's docs | Most follow the same pattern |
The content is identical — only the filename changes. Copy CLAUDE.md, rename it, and you're set.
- Human owns WHAT and WHY. AI owns HOW.
- Every claim needs evidence. No hallucinations.
- Do the task. Only the task. No scope creep.
- A task isn't done until verification passes.