nlp-alignment

Best practices for LLM alignment techniques including RLHF, DPO, and instruction tuning. Use when working on alignment or safety.

Quality

70%

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

Fix and improve this skill with Tessl

tessl review fix ./researchclaw/skills/builtin/domain/nlp-alignment/SKILL.md

Quality

Content

72%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 body is concise and well-organized with specific hyperparameters, but it stops short of executable guidance and lacks validation checkpoints for the risky training workflow. It is a useful quick-reference rather than a complete playbook.

Suggestions

Add a minimal executable example (e.g. a trl DPOTrainer snippet) so the hyperparameters can be applied directly rather than described.

Insert validation checkpoints in the training recipe, such as monitoring reward/KL divergence and halting on reward hacking before proceeding.

Name the concrete library/framework for each method so the recipe is copy-paste ready rather than abstract.

Dimension	Reasoning	Score
Conciseness	Lean bullet points with no padded concept explanations; it assumes Claude knows RLHF/DPO/PPO and every line carries actionable information.	3 / 3
Actionability	Provides concrete hyperparameters (lr=5e-7, beta=0.1, clip=0.2) and named benchmarks (TruthfulQA, BBQ), but gives no runnable code, no library (e.g. trl), and omits key implementation details needed to execute.	2 / 3
Workflow Clarity	A training sequence is present (SFT then DPO/PPO then KL penalty then eval), but there are no validation checkpoints or feedback loops for a fragile RL training process.	2 / 3
Progressive Disclosure	Under 50 lines, single-purpose, and cleanly organized into Methods, Training recipe, and Common pitfalls sections with no need for external references.	3 / 3
	Total	10 / 12 Passed

Description

67%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 covers the alignment domain with an explicit 'Use when' trigger and names specific methods, but it lacks concrete action verbs and leans on the somewhat generic 'safety' term. It is solid but not exemplary.

Suggestions

Replace 'Best practices for' with concrete actions, e.g. 'Train and tune LLMs with RLHF, DPO, and instruction tuning; configure reward models and PPO/KL hyperparameters.'

Broaden trigger terms to include natural variations users say, such as 'preference optimization', 'reward modeling', and 'fine-tuning'.

Narrow the 'safety' trigger or pair it with alignment-specific context to reduce overlap with general safety skills.

Dimension	Reasoning	Score
Specificity	Names the domain and methods ('RLHF, DPO, and instruction tuning') but describes 'best practices for' rather than listing concrete actions, so it is not comprehensive enough for a 3.	2 / 3
Completeness	Explicitly answers both what it does ('Best practices for LLM alignment techniques including RLHF, DPO, and instruction tuning') and when to use it ('Use when working on alignment or safety').	3 / 3
Trigger Term Quality	Includes relevant terms a user in this domain would say ('alignment', 'RLHF', 'DPO', 'safety') but misses common variations like 'preference optimization', 'reward model', or 'fine-tuning'.	2 / 3
Distinctiveness Conflict Risk	The LLM-alignment niche is fairly distinct, but the 'safety' trigger is broad and could overlap with general safety skills, so it is not a clear non-conflicting 3.	2 / 3
	Total	9 / 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/nlp-alignment/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.