rl-policy-optimization

Best practices for reinforcement learning policy optimization. Use when working on RL agents, PPO, SAC, or reward design.

Quality

72%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./researchclaw/skills/builtin/domain/rl-policy-optimization/SKILL.md

Quality

Content

80%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A tight, token-efficient reference of concrete RL training recipes and evaluation protocols with strong actionability and clean section structure. Its main gap is the lack of an explicitly sequenced, checkpointed workflow and the absence of any executable code example.

Suggestions

Add a short numbered workflow (e.g., 1. Select algorithm by action space, 2. Configure hyperparameters from the recipe, 3. Train with vectorized envs, 4. Evaluate over 10+ episodes, 5. Sweep seeds/hyperparameters if unstable) to make the sequence and checkpoints explicit.

Include one minimal copy-paste PPO training snippet (e.g., a Stable-Baselines3 or cleanrl-style config block) to lift actionability from concrete values to fully executable guidance.

Dimension	Reasoning	Score
Conciseness	The body is a lean bullet reference with no padding and no explanation of concepts Claude already knows (e.g., what RL or PPO is); every line delivers a concrete value or directive, matching 'Lean and efficient; assumes Claude's competence'.	5 / 5
Actionability	Concrete, specific guidance throughout (clip=0.2, lr=3e-4, gamma=0.99, GAE lambda=0.95, 10+ eval episodes, 5+ seeds, gymnasium.vector); as an instruction-only skill the absence of code is not penalized, but it stops short of 5 because there is no copy-paste executable example covering common cases.	4 / 5
Workflow Clarity	The four sections imply a loose sequence (select -> train -> evaluate -> pitfalls) and include validation guidance (evaluation protocol), but steps are not explicitly sequenced or checkpointed; this matches 'sequence present but checkpoints missing or implicit' rather than the 4 anchor's clear numbered sequence.	3 / 5
Progressive Disclosure	At ~24 lines with no need for external references, the content is well-organized into four clearly labeled sections (Algorithm selection, Training recipe, Evaluation, Common pitfalls); per the simple-skill exception this scores 5 with just well-organized sections.	5 / 5
	Total	17 / 20 Passed

Description

65%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A concise, third-person description with an explicit 'Use when' trigger clause and strong natural keywords. Its main weakness is the generic 'Best practices for' framing that names the domain without enumerating concrete capabilities.

Suggestions

Replace the generic 'Best practices for' framing with 2-3 concrete capabilities (e.g., 'Tune PPO/SAC hyperparameters, design reward functions, and structure RL training runs').

Add a couple of common synonyms/variants to the trigger clause, such as 'reinforcement learning' and 'policy gradient', to broaden natural keyword coverage.

Dimension	Reasoning	Score
Specificity	The description names the domain ('reinforcement learning policy optimization') but the only action framing is the generic 'Best practices for', with no enumerated concrete capabilities; this matches 'Names the domain but actions are minimal or generic' and falls below the 3 anchor which requires 1-2 concrete actions.	2 / 5
Completeness	Both 'what' ('Best practices for reinforcement learning policy optimization') and 'when' ('Use when working on RL agents, PPO, SAC, or reward design') are present with concrete triggers; not a 5 because the 'what' is a generic best-practices framing rather than a list of concrete actions.	4 / 5
Trigger Term Quality	Natural user-facing terms 'RL agents, PPO, SAC, or reward design' give good keyword coverage; it stops short of 5 because common variations like the full phrase 'reinforcement learning' and 'policy gradient' are absent.	4 / 5
Distinctiveness Conflict Risk	RL policy optimization with PPO/SAC triggers is a clear niche with minimal conflict risk; just short of 5 because 'RL agents' is somewhat broad and could overlap with adjacent RL skills.	4 / 5
	Total	14 / 20 Passed

Validation

100%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository: aiming-lab/AutoResearchClaw
Path: researchclaw/skills/builtin/domain/rl-policy-optimization/SKILL.md
Commit: be4ba47

Reviewed: 5 days ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.