grpo-rl-training

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

Quality

67%

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 ./backend/cli/skills/ml-training/grpo-rl-training/SKILL.md

Quality

Content

77%

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 executable code and a well-sequenced, validated workflow, but it is a long monolithic document that explains some fundamentals Claude already knows and references bundle directories that do not exist. Splitting reference material into actual bundle files and trimming introductory exposition would improve it.

Suggestions

Move the 'Core Concepts / GRPO Algorithm Fundamentals' explanatory material into a separate references file (or remove it) to reduce tokens on concepts Claude already knows.

Either create the referenced 'templates/' and 'examples/' directories with real files, or remove those references to avoid pointing at non-existent bundles.

Tighten the intro paragraph and 'Mathematical Intuition' block, keeping only what is non-obvious for TRL GRPO specifically.

Dimension	Reasoning	Score
Conciseness	The body is mostly efficient with executable code and tables, but it includes explanatory material Claude likely already knows (the 'GRPO Algorithm Fundamentals' section, the mathematical-intuition block, and the 'Critical Difference from PPO' bullets) and could be tightened, matching the 'mostly efficient but includes some unnecessary explanation' anchor.	2 / 3
Actionability	It provides fully executable, copy-paste-ready code for dataset prep, reward functions, GRPOConfig, trainer setup, LoRA/Unsloth, and inference, plus concrete command-style examples, matching the 'fully executable code/commands' anchor.	3 / 3
Workflow Clarity	The four-step implementation workflow is clearly sequenced with validation checkpoints ('Validate data quality before training', 'Test rewards independently'), a before/during/after checklist, and a debugging feedback loop, matching the 'clear sequence with explicit validation steps and checklists' anchor.	3 / 3
Progressive Disclosure	No bundle files exist and the ~560-line body is a monolithic single file with all content inline; it references non-existent 'templates/' and 'examples/' directories, fitting the 'some structure but content that should be separate is inline' anchor rather than a clean overview with one-level-deep references.	2 / 3
	Total	10 / 12 Passed

Description

57%

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 identifies a clear, distinctive niche (GRPO/RL fine-tuning with TRL) but is padded with vague phrasing ('Expert guidance for') and lacks an explicit 'Use when' trigger clause. It answers 'what' reasonably but leaves 'when' only implied.

Suggestions

Replace 'Expert guidance for' with concrete actions, e.g., 'Configure and run GRPO/RL fine-tuning with TRL using custom reward functions'.

Add an explicit trigger clause, e.g., 'Use when fine-tuning LLMs with GRPO/TRL, designing reward functions, or doing RLHF/reinforcement learning training'.

Include more natural trigger terms users would say (reinforcement learning, reward functions, RLHF, PPO) to broaden coverage.

Dimension	Reasoning	Score
Specificity	Names the domain ('GRPO/RL fine-tuning with TRL') and a couple of actions ('reasoning and task-specific model training'), but the leading 'Expert guidance for' is vague and the action list is not comprehensive, matching the anchor that names domain and some actions.	2 / 3
Completeness	It states what the skill provides (GRPO/RL fine-tuning guidance with TRL) but has no explicit 'Use when...' trigger clause, so per the guideline a missing explicit trigger caps completeness at 2.	2 / 3
Trigger Term Quality	Includes relevant natural terms a practitioner would say (GRPO, RL fine-tuning, TRL, reasoning), but misses common variations a user might actually utter such as 'reinforcement learning', 'reward functions', 'RLHF', or 'PPO', fitting the 'some relevant keywords but missing common variations' anchor.	2 / 3
Distinctiveness Conflict Risk	The phrase 'GRPO/RL fine-tuning with TRL' carves out a clear niche with distinct triggers that are unlikely to fire for unrelated skills.	3 / 3
	Total	9 / 12 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 (574 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: synthetic-sciences/openscience
Path: backend/cli/skills/ml-training/grpo-rl-training/SKILL.md
Commit: 52845c3

Reviewed: about 10 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.