rl-policy-optimization

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

Quality

88%

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

Quality

Content

87%Weight 40%Scale 1-3

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

The content is an efficient, actionable reference with exact hyperparameters and a concrete evaluation protocol, well-organized into clear sections. Its main gap is workflow clarity: the implied select-train-evaluate flow lacks explicit validation checkpoints or stopping criteria.

Suggestions

Add an explicit validation/checkpoint step in the training flow, e.g. 'Stop training when eval return plateaus for N updates; verify std across 5+ seeds is within tolerance'.

Sequence the four sections as a numbered workflow (1. Select algorithm -> 2. Train -> 3. Evaluate -> 4. Check pitfalls) so the process order is explicit rather than implied.

Add a brief convergence/sanity-check signal (e.g. expected return range or value-loss behavior) so a run can be verified as healthy rather than just logged.

Dimension	Reasoning	Score
Conciseness	The body is a lean set of bullets with no padding and no explanation of concepts Claude already knows (e.g. what RL or PPO is); every line earns its place, matching the lean-and-efficient anchor.	3 / 3
Actionability	It gives concrete, copy-paste-ready specifics (PPO clip=0.2, lr=3e-4, gamma=0.99, GAE lambda=0.95; SAC tau=0.005, auto-tune alpha; gymnasium.vector; specific metrics to log and a 10+ episode eval protocol); per the rubric's instruction-only note, the absence of code is not penalized because the guidance is actionable.	3 / 3
Workflow Clarity	The sections imply a loose sequence (Algorithm selection -> Training recipe -> Evaluation -> Common pitfalls) but there are no explicit validation checkpoints or feedback loops (e.g. when to stop training, how to confirm a run is stable); not 1 because steps are present and ordered, not 3 because checkpoints are missing/implicit.	2 / 3
Progressive Disclosure	The body is under 50 lines with no need for external references and is organized into four clearly labeled sections; per the simple-skills note this scores 3 with just well-organized sections. No bundle files exist to verify against.	3 / 3
	Total	11 / 12 Passed

Description

90%Weight 40%Scale 1-3

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

The description is concise, uses third person, and pairs a clear domain statement with an explicit 'Use when' trigger clause naming concrete algorithms. Its only weakness is that it describes the topic as 'best practices' rather than enumerating the concrete actions the skill performs.

Dimension	Reasoning	Score
Specificity	"Best practices for reinforcement learning policy optimization" names the domain and surfaces PPO/SAC, but it describes a topic rather than listing multiple concrete actions (e.g. tune hyperparameters, design reward functions, train agents), so it lands at the anchor that names domain and some actions without being comprehensive.	2 / 3
Completeness	It states what the skill covers ("Best practices for reinforcement learning policy optimization") and an explicit when ("Use when working on RL agents, PPO, SAC, or reward design"), clearly answering both what and when with an explicit trigger clause.	3 / 3
Trigger Term Quality	"Use when working on RL agents, PPO, SAC, or reward design" plus "reinforcement learning" in the opening gives good coverage of natural terms a user would actually say; not the level below because it is not missing common variations.	3 / 3
Distinctiveness Conflict Risk	Reinforcement learning policy optimization is a clear niche and the named algorithm triggers (PPO, SAC) make it unlikely to fire for unrelated skills; not 2 because the triggers are specific rather than broadly overlapping.	3 / 3
	Total	11 / 12 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: e2e23c9

Reviewed: 6 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.