Name: rl-policy-optimization
Author: raja21068

Imported from raja21068/AutoScholar (skills/builtin/domain/rl-policy-optimization/SKILL.md). Install upstream with npx skills add raja21068/AutoScholar --skill rl-policy-optimization. Copyright stays with the author.

RL Policy Optimization Best Practice

Algorithm selection:

Discrete actions: PPO, DQN, A2C
Continuous actions: SAC, TD3, PPO
Multi-agent: MAPPO, QMIX
Offline: CQL, IQL, Decision Transformer

Training recipe:

PPO: clip=0.2, lr=3e-4, gamma=0.99, GAE lambda=0.95
SAC: lr=3e-4, tau=0.005, auto-tune alpha
Use vectorized environments (e.g., gymnasium.vector)
Normalize observations and rewards
Log episode return, episode length, value loss, policy entropy

Evaluation:

Report mean +/- std over 10+ evaluation episodes
Use deterministic policy for evaluation
Compare against random policy and simple baselines
Report sample efficiency (return vs. env steps)

Common pitfalls:

Reward shaping can introduce bias
Seed sensitivity is HIGH — use 5+ seeds
Hyperparameter sensitivity — do a small sweep

{ "ocm": "1", "id": "raja21068-autoscholar-rl-policy-optimization", "kind": "skill", "name": "rl-policy-optimization", "description": "Best practices for reinforcement learning policy optimization. Use when working on RL agents, PPO, SAC, or reward design.", "publisher": "raja21068", "version": "1.0.0", "capabilities": { "domains": [ "general" ], "tags": [ "skill-md", "github" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "Best practices for reinforcement learning policy optimization. Use when working on RL agents, PPO, SAC, or reward design." ], "primary": false, "metadata": { "source": { "provider": "github", "repository": "https://github.com/raja21068/AutoScholar", "path": "skills/builtin/domain/rl-policy-optimization/SKILL.md", "ref": "88bb3f5bd26c4b1f1f7e35879a02ad67a2b6f7d3", "url": "https://github.com/raja21068/AutoScholar/blob/88bb3f5bd26c4b1f1f7e35879a02ad67a2b6f7d3/skills/builtin/domain/rl-policy-optimization/SKILL.md", "key": "raja21068/AutoScholar/skills/builtin/domain/rl-policy-optimization/SKILL.md" } }, "instructions": "## RL Policy Optimization Best Practice\nAlgorithm selection:\n- Discrete actions: PPO, DQN, A2C\n- Continuous actions: SAC, TD3, PPO\n- Multi-agent: MAPPO, QMIX\n- Offline: CQL, IQL, Decision Transformer\n\nTraining recipe:\n- PPO: clip=0.2, lr=3e-4, gamma=0.99, GAE lambda=0.95\n- SAC: lr=3e-4, tau=0.005, auto-tune alpha\n- Use vectorized environments (e.g., gymnasium.vector)\n- Normalize observations and rewards\n- Log episode return, episode length, value loss, policy entropy\n\nEvaluation:\n- Report mean +/- std over 10+ evaluation episodes\n- Use deterministic policy for evaluation\n- Compare against rand", "cost": { "context_tokens": 206 } }

rl-policy-optimization

About

RL Policy Optimization Best Practice

Use it

Manifest

Reviews