spinning-up-deep-rl

Knowledge base from "Spinning Up in Deep RL" by Joshua Achiam (OpenAI, MIT-licensed). Use when applying Achiam's frameworks for RL fundamentals and MDPs, the model-free algorithm taxonomy, policy gradient derivations, the six reference algorithms (VPG, TRPO, PPO, DDPG, TD3, SAC), debugging silently-failing RL code, or running rigorous multi-seed RL experiments.

Quality

85%

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

71%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 well-structured, dense knowledge-base skill with clear invocation guidance and strong conceptual reference material. Its main weakness is bundle integrity: the body references 20 chapter files plus glossary/patterns/cheatsheet, none of which are present, breaking the progressive-disclosure navigation it promises.

Suggestions

Ship the referenced bundle files (chapters/ch01..ch20, glossary.md, patterns.md, cheatsheet.md) or remove the dangling links — currently every chapter and supporting-file link resolves to nothing.

Trim the Core Frameworks entries that restate general RL fundamentals Claude already knows (e.g., Bellman self-consistency, the five Phi choices) to keep only Achiam's specific synthesized framing.

Make the scope/boundary guardrail an explicit workflow checkpoint (e.g., 'If a topic is outside Core Frameworks, read the chapter file before answering; if outside the source's scope, state that explicitly') so the validation step is signaled rather than implied.

Dimension	Reasoning	Score
Conciseness	Dense and notation-driven with no hand-holding ('pi* = argmax_pi J(pi)', the EGLP lemma, the squeeze-bug archetype), assuming Claude's competence; however some Core Frameworks restate general RL fundamentals (Bellman self-consistency, the five Phi choices) Claude already knows and could be trimmed. Not a 3 because nothing is padded or verbose; not a 5 because of the restated known concepts.	4 / 5
Actionability	Concrete invocation examples ('/cs:spinning-up-deep-rl entropy regularization', 'ch09') and a clear topic-resolution path plus specific debugging steps ('Debug by measuring everything', the [N] vs [N,1] squeeze archetype); minor gaps only — no executable code, but this is an instruction/knowledge skill so that absence is not penalized.	4 / 5
Workflow Clarity	The 'How to Use' section gives a clear sequenced decision tree (no argument -> core; a topic -> Topic Index -> chapter; chNN -> summary; 'what chapters' -> index) with the guardrail 'read the relevant chapter file before answering rather than guessing'. Not a 5 because validation/checkpoint feedback loops are only implicit (the Scope & Limits 'say so rather than improvising' boundary), and not a 3 because the sequence is explicit and unambiguous.	4 / 5
Progressive Disclosure	The in-body structure is well organized with a clear overview, Chapter Index, Topic Index, and clearly signaled one-level-deep markdown links; however every referenced bundle path is missing — no chapters/ directory and no glossary.md/patterns.md/cheatsheet.md exist — so the navigation points to nothing. Not a 4 because the empty bundle is more than a 'minor organization gap'; not a 2 because the textual structure and signaling are genuinely good, references are not buried.	3 / 5
	Total	15 / 20 Passed

Description

100%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 strong description: it names a specific source and author, enumerates concrete capabilities with the actual algorithm vocabulary, and gives an explicit 'Use when' clause with multiple concrete triggers. It reads in third person and avoids fluff or over-claims.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete capabilities — 'RL fundamentals and MDPs, the model-free algorithm taxonomy, policy gradient derivations, the six reference algorithms (VPG, TRPO, PPO, DDPG, TD3, SAC), debugging silently-failing RL code, or running rigorous multi-seed RL experiments' — comprehensive and specific rather than generic.	5 / 5
Completeness	Explicitly states both what ('Knowledge base from Spinning Up in Deep RL by Joshua Achiam') and when ('Use when applying Achiam's frameworks... debugging... running rigorous multi-seed RL experiments'), matching the anchor's concrete-trigger-phrases pattern.	5 / 5
Trigger Term Quality	Covers the natural terms practitioners actually say — named algorithms (VPG, TRPO, PPO, DDPG, TD3, SAC), 'policy gradient', 'MDPs', 'debugging silently-failing RL code', 'multi-seed RL experiments' — comprehensive coverage of the niche's vocabulary.	5 / 5
Distinctiveness Conflict Risk	Occupies a clear niche (Achiam's Spinning Up frameworks specifically) with distinct triggers — named algorithms and named author — minimizing overlap with generic RL or coding skills.	5 / 5
	Total	20 / 20 Passed

Validation

93%

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 — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
relative_links	Relative link issues: 23 missing	Warning

	Total	15 / 16 Passed

Repository: alirezarezvani/claude-skills
Path: engineering/spinning-up-deep-rl/skills/spinning-up-deep-rl/SKILL.md
Commit: 19392f7

Reviewed: ~2 hours 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.