CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

stable-baselines3

Production-ready reinforcement learning algorithms (PPO, SAC, DQN, TD3, DDPG, A2C) with scikit-learn-like API. Use for standard RL experiments, quick prototyping, and well-documented algorithm implementations. Best for single-agent RL with Gymnasium environments. For high-performance parallel training, multi-agent systems, or custom vectorized environments, use pufferlib instead.

Quality

83%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

75%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 highly actionable, well-structured body: executable code throughout, a sequenced workflow with an explicit validation step, and clean one-level-deep references to real bundle files. The main costs are token efficiency — overview/purpose prose and inlined advanced detail that could move to references — and a missing re-validation feedback loop in the workflow.

Suggestions

Trim the Overview paragraph and the per-section 'Purpose:' statements (e.g., 'Callbacks enable monitoring metrics, saving checkpoints...') — Claude already knows what SB3, callbacks, and vectorized environments are; keep only the non-obvious constraints.

Move advanced inlined detail (the full custom callback class, HER replay-buffer configuration, video recording, and multi-input policies) into the existing references (callbacks.md, custom_environments.md) and leave one-line pointers, shrinking SKILL.md to a true overview.

Add an explicit feedback loop to the workflow after the check_env() step, e.g., 'If check_env reports errors, fix the environment and re-run until it passes before training,' and reference the Common Issues section for recovery.

DimensionReasoningScore

Conciseness

The code-first sections are dense and useful, but the Overview paragraph ('SB3 is a PyTorch-based library providing reliable implementations...') and per-section 'Purpose:' statements re-explain concepts Claude already knows, matching the 'mostly efficient but includes some unnecessary explanation' anchor rather than the minor-trim anchor 4.

3 / 5

Actionability

Nearly every section contains copy-paste-ready executable snippets covering the common cases (training, custom envs, vectorized envs, callbacks, save/load, evaluation, HER, TensorBoard), plus concrete pitfalls like 'the replay buffer is NOT saved' and the Discrete start!=0 restriction — a clear anchor 5.

5 / 5

Workflow Clarity

The 8-step workflow has a clear sequence, an explicit validation checkpoint ('Always run check_env() before training'), and a Common Issues troubleshooting section, but lacks an explicit validate-fix-revalidate loop within the workflow itself, fitting anchor 4 rather than 5.

4 / 5

Progressive Disclosure

Every capability section points to a real one-level-deep reference or script (all listed bundle files exist) and a Resources section indexes them, but ~290 lines of inlined detail (full callback class, HER configuration, video recording) that belongs in the references keeps it at the 'good structure, most content appropriately placed' anchor rather than 5.

4 / 5

Total

16

/

20

Passed

Description

87%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: concrete, third-person, with explicit use and do-not-use triggers and a clear boundary against a competing skill. Its only weakness is that it names the library's algorithms rather than the actions the skill itself performs, leaving minor specificity and trigger-synonym gaps.

DimensionReasoningScore

Specificity

Names six concrete algorithms (PPO, SAC, DQN, TD3, DDPG, A2C), the scikit-learn-like API, and the Gymnasium environment framework, but describes the library rather than enumerating the actions the skill performs (e.g., train agents, create custom environments), so it falls just short of the comprehensive anchor 5.

4 / 5

Completeness

Explicitly answers both 'what' (production-ready RL algorithms with a scikit-learn-like API) and 'when' ('Use for standard RL experiments, quick prototyping... Best for single-agent RL with Gymnasium environments'), with concrete trigger phrases and explicit anti-triggers pointing to pufferlib — a clear anchor 5 match.

5 / 5

Trigger Term Quality

Good natural-term coverage ('reinforcement learning', 'RL experiments', 'Gymnasium', algorithm names, and negative triggers like 'multi-agent systems'), but a few common user phrases such as 'train an RL agent' or 'Gym' are missing, so it fits anchor 4 rather than 5.

4 / 5

Distinctiveness Conflict Risk

Clear niche (single-agent RL with Gymnasium via SB3) with explicit boundary guidance routing multi-agent, parallel, and custom-vectorized cases to pufferlib, giving minimal conflict risk — matches anchor 5.

5 / 5

Total

18

/

20

Passed

Validation

87%

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

Validation — 14 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

metadata_version

'metadata.version' is missing

Warning

frontmatter_unknown_keys

Unknown frontmatter key(s) found; consider removing or moving to metadata

Warning

Total

14

/

16

Passed

Repository

synthetic-sciences/openscience

Path

backend/cli/skills/ml-training/stable-baselines3/SKILL.md

Reviewed

Table of Contents

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.