rl-expert-skills

45 reinforcement learning skills for Claude Code and OpenAI Codex.

Covers architecture, algorithms, environments, evaluation, and MLOps.

Usage

From any project folder:

npx rl-expert-skills

The installer will ask:

1. Which tool? — Claude Code, OpenAI Codex, or both
2. Scope? — current project or current user

What gets installed

| Tool | Project scope | User scope | |------|--------------|------------| | Claude Code | .claude/skills/ | ~/.claude/skills/ | | OpenAI Codex | .codex/skills/ + AGENTS.md | ~/.codex/skills/ + ~/.codex/instructions.md |

Skills included

Orchestrator agents

• rl-architect — MDP formulation, algorithm selection, system blueprint
• rl-algo-engineer — neural networks, training loop, loss functions
• rl-env-engineer — Gymnasium/PettingZoo environments and wrappers
• rl-evaluator — evaluation, OOD testing, failure diagnosis
• rl-mlops-engineer — experiment tracking, distributed training, checkpointing

Individual skills (40 skills covering reward design, policy/critic architecture, MARL, offline RL, RLHF, domain randomization, and more)

Publishing

First time

cd path/to/rl-expert-skills
npm login
npm publish --access public

Updating skills

After editing the source skills, re-bundle and republish:

cd path/to/rl-expert-skills
npm run prepare       # re-copies skills from ../RL-Expert/skills/
npm version patch     # bump version (patch / minor / major)
npm publish --access public

Using before publishing (local test)

From any project folder:

npx path/to/rl-expert-skills