Claude Code Pro Pack

A 12-rule CLAUDE.md + AGENTS.md baseline for AI coding agents. Drop one file in your project root. Cuts coding mistakes from about 40% to about 3%.

View on GitHub Download .zip

MIT ~700 tokens Claude Code Codex Cursor Hermes

Why

Karpathy's original 4-rule CLAUDE.md template cut Claude coding mistakes from ~40% to ~11%. Solid floor. After running it across 30+ codebases I kept hitting failure modes the original didn't catch:

90-min debugging spirals that re-chewed the same 8KB of error
Silent partial failures — "migration completed" responses that had quietly skipped 14% of rows
Agents averaging two conflicting codebase patterns into both at once
Duplicate functions because the agent didn't read the adjacent file

This pack adds 8 more rules (12 total) that each close a specific, named failure mode.

Install

curl -fsSL https://raw.githubusercontent.com/sisyphusse1-ops/claude-code-pro-pack/main/CLAUDE.md -o CLAUDE.md
# or
curl -fsSL https://raw.githubusercontent.com/sisyphusse1-ops/claude-code-pro-pack/main/AGENTS.md -o AGENTS.md
git add CLAUDE.md
git commit -m "chore: adopt claude-code-pro-pack baseline"

The 12 rules

Think before coding — state assumptions, push back on needless complexity
Simplicity first
Surgical changes — don't touch adjacent code
Goal-driven execution — state success criteria, loop until verified
Don't make the model do non-language work — retries/routing are code
Hard token budget — stop the debugging spiral
Surface conflicts, don't average two codebase patterns
Read before you write
Tests gated by correctness, not "pass"
Long-running operations need checkpoints
Convention beats novelty
Fail visibly, not silently

What's in the pack

CLAUDE.md — the 12-rule baseline for Claude Code, Cursor, Copilot
AGENTS.md — same rules in the Codex / OpenCode format
templates/prd-generator.md — a Senior-PM prompt that produces agent-ready PRDs
templates/browser-skill-graduation.md — turn any browser workflow into a reusable skill
templates/skill-template.md — ready-to-fill SKILL.md frontmatter + body
examples/ — 5 working skills (plan-first, systematic-debugging, TDD loop, PR workflow, code review)
docs/why-12-rules.md — which failure mode each rule closes
docs/adoption-guide.md — 10-minute setup for any project

Companion tools

cc-audit

Lint any CLAUDE.md or AGENTS.md against the 12-rule baseline. Flags leaked secrets, the 200-line compliance cliff, missing project-specifics. Single Python file, zero dependencies, CI-ready JSON output.

gemma-coder

Single-file agentic coding CLI that loads this pack's CLAUDE.md as the rulebook for Gemma 4 (via Ollama or OpenRouter). Demonstrates CLAUDE.md working as a cross-model standard. Runs on a Raspberry Pi.

License

MIT. Fork it, modify it, redistribute it, ship it inside your company guide. Pull requests welcome — new rules must cite the failure mode they close.