prompting-buddy

A Claude Code skill that writes your CLAUDE.md, .claude/rules/, and .claude/agents/ files for you — grounded in Anthropic's official best practices.

You describe your project. Prompting Buddy scans your repo, asks a few sharp questions, and writes prompt files directly into your project. No copy-paste, no guesswork about what belongs in CLAUDE.md vs rules vs agents.

What it does

1. Fetches the latest official docs from Anthropic before writing anything — so the advice never goes stale
2. Scans your repo to detect tech stack, build tools, linting config, CI setup, and existing prompt files
3. Asks targeted questions to fill gaps the code doesn't reveal (workflow, conventions, team quirks)
4. Generates prompt files — CLAUDE.md under 200 lines, path-scoped rules, properly formatted agent definitions
5. Self-reviews against a checklist before writing: no codebase tours, no teaching basics, no bloat
6. Writes files directly into your project — or falls back to markdown blocks if you prefer

It also reviews and optimizes existing CLAUDE.md files — trimming redundant emphasis, removing self-evident rules, and restructuring for better adherence.

Install

Download the latest prompting-buddy.skill from Releases, then in Claude Code:

/skill install path/to/prompting-buddy.skill

Or clone this repo into your skills directory:

git clone https://github.com/YOUR_USERNAME/prompting-buddy.git ~/.claude/skills/prompting-buddy

Usage

Once installed, the skill triggers automatically when you ask Claude Code things like:

• "Set up CLAUDE.md for this project"
• "I need rules files scoped to each service in my monorepo"
• "Create a code-review agent"
• "My CLAUDE.md is too long and Claude keeps ignoring half of it — fix it"
• "Migrate my .cursorrules to Claude Code"

Or invoke it directly:

/prompting-buddy set up Claude Code for this project

Example: monorepo with 3 services

> I have a Python monorepo with api-gateway (FastAPI), ml-pipeline (PyTorch),
> and data-ingestion (Airflow). Each has different conventions. Set me up.

Prompting Buddy generates:

Created:
  ./CLAUDE.md                              (78 lines)
  ./.claude/rules/api-gateway.md           (24 lines, paths: services/api-gateway/**)
  ./.claude/rules/ml-pipeline.md           (18 lines, paths: services/ml-pipeline/**)
  ./.claude/rules/data-ingestion.md        (20 lines, paths: services/data-ingestion/**)
  ./.claude/agents/code-reviewer.md        (32 lines, model: sonnet, read-only)

Contributing

The best way to improve this skill is to run evals, review the outputs, and iterate on SKILL.md. If Claude generates bad prompt files for a particular project type, add it as a new eval case in evals/evals.json and use it to drive the fix.

License

MIT