grpo-rl-training

Expert guidance for GRPO/RL fine-tuning with TRL for reasoning and task-specific model training

Quality

59%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Critical

Do not install without reviewing

Fix and improve this skill with Tessl

tessl review fix ./skills/post-training/grpo-rl-training/SKILL.md

Quality

Content

65%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.

The content is highly actionable with comprehensive executable examples, but it is over-long for an overview, lacks mid-workflow validation gates for a batch training operation, and claims bundle directories (templates/, examples/) that are absent.

Suggestions

Move the conceptual primers (algorithm fundamentals, math intuition, when-NOT-to-use) and the 'Usage Instructions for Agents' recap into a separate reference file or trim them to keep SKILL.md a lean overview.

Insert explicit validation checkpoints in the main workflow (e.g., 'Test reward functions on sample completions before full training; only proceed when rewards are non-zero and varied').

Either create the referenced 'templates/' and 'examples/' directories with the corresponding files, or remove the dangling references so navigation is not broken.

Dimension	Reasoning	Score
Conciseness	The ~560-line body is mostly dense and useful, but includes redundant conceptual primers (algorithm fundamentals, math intuition, when-NOT-to-use) and a recap-style 'Usage Instructions for Agents' section that an expert-level skill targeting Claude could trim.	3 / 5
Actionability	Provides fully executable, copy-paste-ready code across the whole workflow (dataset prep, three reward functions, two GRPOConfigs, LoRA and Unsloth setup, merge and inference) plus parameter tables covering common cases.	5 / 5
Workflow Clarity	The Step 1-4 sequence and Before/During/After checklists give a clear path, but for a long-running batch/training operation the main flow lacks explicit validate-then-proceed gates; per the feedback-loops guidance this caps at 3.	3 / 5
Progressive Disclosure	Header organization is decent, but the body is a monolithic ~560-line file that inlines content arguably belonging in separate files, and it references 'templates/' and 'examples/' directories that do not exist in the bundle.	3 / 5
	Total	14 / 20 Passed

Description

53%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.

The description is specific and reasonably distinct, but it lacks an explicit 'when to use' trigger clause and leans on technical jargon over natural user phrases. It is competent but not exemplary.

Suggestions

Add an explicit 'Use when ...' clause naming concrete triggers (e.g., 'Use when fine-tuning a model with custom reward functions, enforcing structured output formats, or teaching verifiable reasoning tasks').

Broaden trigger terms with natural synonyms users say: 'RLHF', 'PPO', 'DPO', 'reward function', 'reinforcement learning fine-tuning'.

Enumerate a few more concrete actions (e.g., 'design reward functions, configure GRPO training, monitor reward/KL metrics') to lift specificity.

Dimension	Reasoning	Score
Specificity	Names the domain (GRPO/RL fine-tuning, TRL) and 2-3 actions (fine-tuning, reasoning, task-specific training), but the actions are abstract categories rather than several distinct concrete operations.	3 / 5
Completeness	Has a clear 'what' but no explicit 'Use when...' trigger clause; per the rubric, a missing explicit trigger guidance caps completeness at 3.	3 / 5
Trigger Term Quality	Includes relevant technical keywords (GRPO, RL fine-tuning, TRL) but omits natural conversational phrases and common synonyms (RLHF, PPO, reward functions) a user would actually say.	3 / 5
Distinctiveness Conflict Risk	The GRPO/RL-with-TRL niche is mostly distinct, with only minor overlap risk against sibling post-training skills like DPO/PPO/RLHF.	4 / 5
	Total	13 / 20 Passed

Validation

87%

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 — 14 / 16 Passed

Validation for skill structure

Criteria	Description	Result
skill_md_line_count	SKILL.md is long (573 lines); consider splitting into references/ and linking	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	14 / 16 Passed

Repository: OpenLAIR/dr-claw
Path: skills/post-training/grpo-rl-training/SKILL.md
Commit: 4b35756

Reviewed: 1 day 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.