hugging-face-evaluation

Add and manage evaluation results in Hugging Face model cards. Supports extracting eval tables from README content, importing scores from Artificial Analysis API, and running custom model evaluations with vLLM/lighteval. Works with the model-index metadata format.

Quality

68%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Fix and improve this skill with Tessl

tessl review fix ./backend/cli/skills/ml-training/hugging-face-evaluation/SKILL.md

Quality

Content

77%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 highly actionable with clear, validated workflows, but it suffers from heavy internal duplication and a monolithic structure that should be split into reference files. A stale reference to a non-existent train_sft_example.py script slightly undermines otherwise copy-paste-ready guidance.

Suggestions

De-duplicate the workflow: keep one authoritative sequence (get-prs → inspect-tables → extract-readme → apply/create-pr) and reference it from other sections instead of repeating the full commands in four places.

Move the long Commands Reference, Troubleshooting, lighteval task-format catalog, and Integration Examples into separate reference files under a references/ directory, keeping SKILL.md a concise overview with clearly signaled one-level-deep links.

Remove or fix the broken reference to scripts/train_sft_example.py (not present in scripts/), and either document or delete the unreferenced test_extraction.py.

Dimension	Reasoning	Score
Conciseness	The body is mostly efficient and actionable but heavily duplicated — the extract-readme/get-prs workflow recurs in 'Core Capabilities', 'Method 1', 'Commands Reference', and 'Common Patterns', and dependency lists repeat the frontmatter — so it could be tightened substantially.	2 / 3
Actionability	It provides many concrete, copy-paste-ready bash commands with real flags, model IDs, and task strings (e.g. `uv run scripts/lighteval_vllm_uv.py --model meta-llama/Llama-3.2-1B --tasks "leaderboard\|mmlu\|5"`), matching the fully-executable anchor; the one stale reference to a non-existent train_sft_example.py is a minor blemish rather than a systemic gap.	3 / 3
Workflow Clarity	The key workflow is explicitly sequenced (get-prs → inspect-tables → extract-readme → --apply/--create-pr) with a validation checklist, preview-before-apply default behavior, and an explicit feedback loop (check open PRs, warn the user, show URLs, only proceed on confirmation) for the destructive PR-creation operation.	3 / 3
Progressive Disclosure	Sections are well-organized and the provided scripts are referenced correctly, but the SKILL.md is a ~650-line monolith with no references/ directory — detailed troubleshooting, the lighteval task-format catalog, the commands reference, and integration examples that belong in separate files are all inline — matching the score-2 anchor of structure present but content that should be separate kept inline.	2 / 3
	Total	10 / 12 Passed

Description

60%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 is specific and action-oriented, clearly stating what the skill does across three methods. Its main weakness is the missing "Use when..." trigger guidance and limited natural keyword coverage, which caps completeness and distinctiveness at 2.

Suggestions

Add a 'Use when...' clause naming natural trigger terms, e.g. 'Use when adding benchmark or evaluation results to a Hugging Face model card, importing Artificial Analysis scores, or running MMLU/GSM8K evaluations with vLLM or lighteval.'

Include common user-facing keywords such as 'benchmark', 'leaderboard', and 'model card metadata' alongside the existing technical terms.

Sharpen distinctiveness by foregrounding the unique combination (HF model-index metadata + Artificial Analysis + vLLM/lighteval) so it does not collide with generic Hugging Face skills.

Dimension	Reasoning	Score
Specificity	The description lists multiple concrete actions — "extracting eval tables from README content", "importing scores from Artificial Analysis API", and "running custom model evaluations with vLLM/lighteval" — matching the score-3 anchor of several specific concrete actions.	3 / 3
Completeness	The "what" is clearly answered, but there is no "Use when..." clause or equivalent explicit trigger guidance, so per the rubric guideline completeness is capped at 2.	2 / 3
Trigger Term Quality	It includes relevant domain terms ("Hugging Face model cards", "evaluation results", "eval tables", "scores") but omits common variations a user would naturally say ("benchmark", "leaderboard"), and leans on technical jargon ("Artificial Analysis", "model-index metadata").	2 / 3
Distinctiveness Conflict Risk	The niche (evaluation results on HF model cards via specific APIs/frameworks) is fairly specific, but the absence of explicit triggers and the broad "Hugging Face" scope means it could overlap with general HF or model-training skills.	2 / 3
	Total	9 / 12 Passed

Validation

75%

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 — 12 / 16 Passed

Validation for skill structure

Criteria	Description	Result
skill_md_line_count	SKILL.md is long (658 lines); consider splitting into references/ and linking	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning
relative_links	Relative link issues: 1 suspicious	Warning
referenced_paths_exist	Referenced path issues: 1 missing	Warning

	Total	12 / 16 Passed

Repository: synthetic-sciences/openscience
Path: backend/cli/skills/ml-training/hugging-face-evaluation/SKILL.md
Commit: dc00e07

Reviewed: 7 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.