clip

Zero-shot image classification and image-text search.

Quality

53%

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 ./optional-skills/mlops/clip/SKILL.md

Quality

Content

53%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 well-organized recipe collection with mostly executable code, undermined by unsignaled and duplicated bundle content, missing validation checkpoints for batch operations, and padded promotional/background material. The orphaned references/applications.md is the largest structural defect.

Suggestions

Link references/applications.md from the body (e.g., an '## Applications' section pointing to it) and remove the duplicated recipes so each lives in exactly one place.

Add validation/verification guidance to the batch processing and vector-DB sections (e.g., verify embedding shapes, check retrieval sanity, confirm normalized embeddings before cosine similarity).

Trim the Metrics/Resources sections (GitHub star counts, license repetition) to content Claude cannot already know or that does not aid execution.

Dimension	Reasoning	Score
Conciseness	The code recipes are lean, but the body pads with promotional metrics ("25,300+ GitHub stars", "Matches ResNet-50 on ImageNet"), performance tables, and general CLIP background Claude already knows, and it duplicates recipes that appear in references/applications.md. Fits anchor 3 (mostly efficient but includes unnecessary sections that could be trimmed); not 4 because the Metrics and star-count content is clearly dispensable.	3 / 5
Actionability	Mostly executable, concrete snippets covering classification, search, moderation, batching, and vector-DB integration, with copy-paste-ready structure. Falls short of anchor 5 due to minor gaps: the Chroma section reuses undefined variables (image_paths, image_embeddings) and queries with an unnormalized text_embedding on the wrong device.	4 / 5
Workflow Clarity	Each recipe is internally sequenced, but there are no validation or verification checkpoints for the batch-processing and vector-DB indexing workflows, and no error-recovery guidance. The rubric's guideline caps workflow clarity at 3 when batch operations lack validation, so it cannot score 4.	3 / 5
Progressive Disclosure	The bundle provides references/applications.md, but the body never links to or mentions it — the reference is orphaned and undiscoverable — while recipe content that duplicates it (classification, search, moderation, best practices) is inlined in SKILL.md. Matches anchor 2 (references buried; content that belongs in a separate file is inlined); not 3 because the body has good section headers yet the sole reference file is completely unsignaled.	2 / 5
	Total	12 / 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.

A concise, specific description that states what the skill does but never says when to use it. Adding a 'Use when...' clause with trigger terms like CLIP, image embeddings, and image-text similarity would raise completeness and trigger quality substantially.

Suggestions

Append a trigger clause such as: 'Use when the user mentions CLIP, zero-shot image classification, image-text similarity, semantic image search, or image embeddings.'

Include the model name 'CLIP' and synonyms like 'image search' and 'cross-modal retrieval' so natural user phrasings match.

Mention 1-2 more concrete capabilities (e.g., content moderation, image deduplication) to round out coverage.

Dimension	Reasoning	Score
Specificity	"Zero-shot image classification and image-text search" names the vision-language domain plus exactly two concrete actions, matching anchor 3 (1-2 concrete actions, not comprehensive). Not 4 because other capabilities the body covers (cross-modal retrieval, content moderation, embedding computation) are omitted from the description.	3 / 5
Completeness	The 'what' is clear (zero-shot image classification, image-text search) but there is no 'Use when...' clause or equivalent trigger guidance, which the rubric explicitly caps at 3. Not 2 because the 'what' half is concrete rather than vague.	3 / 5
Trigger Term Quality	"image classification" and "image search" are phrases users would naturally say, but the description omits the model name "CLIP" and common synonyms like image similarity, photo tagging, or image embeddings. Fits anchor 3 (some relevant keywords, missing common variations); not 4 because more than a few natural terms are missing.	3 / 5
Distinctiveness Conflict Risk	"Zero-shot image classification and image-text search" carves a specific vision-language niche with only minor overlap risk against related skills (captioning, vision chat, general image tools). Matches anchor 4; not 5 because without 'CLIP' or explicit triggers it could still be confused with broader multimodal skills.	4 / 5
	Total	13 / 20 Passed

Validation

81%

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 — 13 / 16 Passed

Validation for skill structure

Criteria	Description	Result
metadata_version	'metadata.version' is missing	Warning
metadata_field	'metadata' should map string keys to string values	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	13 / 16 Passed

Repository: NousResearch/hermes-agent
Path: optional-skills/mlops/clip/SKILL.md
Commit: 68fa7e9

Reviewed: 3 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.