CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

sentence-transformers

Framework for state-of-the-art sentence, text, and image embeddings. Provides 5000+ pre-trained models for semantic similarity, clustering, and retrieval. Supports multilingual, domain-specific, and multimodal models. Use for generating embeddings for RAG, semantic search, or similarity tasks. Best for production embedding generation.

Quality

77%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./backend/cli/skills/llm-tools/sentence-transformers/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

71%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-structured, largely actionable reference for the library's core use cases with executable quick-start and integration examples. The main weaknesses are the orphaned, unreferenced bundle file whose content is duplicated inline, a few non-executable code snippets, and light marketing padding.

Suggestions

Link references/models.md from the body (e.g., '## Popular models — See [models.md](references/models.md) for the full catalog and benchmarks') and remove the duplicated model-selection and performance tables from SKILL.md.

Fix non-executable snippets: replace the '...' in the batch-encoding example with a real list, and define or explain the variables (embedding1/embedding2, docs, documents) used in the similarity and integration examples.

Trim the 'Metrics' bullet list and star counts (marketing fluff Claude doesn't need) and add a brief validation step to fine-tuning (e.g., evaluate on a held-out set before model.save()).

DimensionReasoningScore

Conciseness

The body is mostly lean: short section headers followed by executable code with minimal prose, assuming Claude's competence ('Load model', 'Encode corpus'). It earns only 4 because of padded marketing content ('15,700+ GitHub stars', 'Metrics' list) and duplication — the popular-models catalog, model-selection table, and performance table all repeat material already in references/models.md.

4 / 5

Actionability

Most sections give copy-paste-ready code (installation, basic usage, semantic search, fine-tuning, LangChain/LlamaIndex). It stops short of 5 because several snippets are not executable as written: 'sentences = ["sentence 1", "sentence 2", ...] * 1000' uses invalid Python ellipsis, and the similarity/integration snippets reference undefined variables (embedding1, embedding2, docs, documents).

4 / 5

Workflow Clarity

The quick start is an unambiguous single-action flow (install → load → encode → compare), and the fine-tuning section has a clear data → dataloader → loss → fit → save sequence. It misses 5 because there are no validation checkpoints or feedback loops (e.g., evaluating a fine-tuned model before saving, verifying embeddings are normalized when that practice is recommended), leaving minor validation gaps.

4 / 5

Progressive Disclosure

The body has well-organized sections, but the bundle's one reference file (references/models.md) is never mentioned or linked anywhere in SKILL.md, so references are not signaled at all — and the body's model catalog, selection matrix, and performance tables duplicate that file's content inline. This matches the anchor 'references present but not clearly signaled; content that should be separate is inline' rather than the well-signaled structure of 4.

3 / 5

Total

15

/

20

Passed

Description

75%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 strong description with a clear what, an explicit 'Use for' trigger clause, and good natural keyword coverage. It reads slightly above the midpoint overall: specific and distinct, but the trigger phrasing is task-based rather than user-utterance-based and a few common synonyms (e.g., 'vector search') are absent.

Suggestions

Rewrite the trigger clause to be user-utterance oriented, e.g. 'Use when the user mentions embeddings, semantic similarity, vector search, RAG retrieval, or wants to run models locally without an API.'

Add the differentiator that distinguishes it from competing embedding skills (local/no-API, free) to the description itself rather than only the body.

Include synonyms like 'vector search' and 'text/image embeddings' as explicit trigger terms.

DimensionReasoningScore

Specificity

The description lists several concrete capabilities — 'semantic similarity, clustering, and retrieval', 'generating embeddings for RAG, semantic search' — grounded in a named framework, matching the 'several specific actions; minor gaps' anchor. It falls short of 5 because actions are task categories rather than a comprehensive action list, and short of 'Best for production embedding generation' being purely a positioning claim rather than an action.

4 / 5

Completeness

Both are present: the 'what' is explicit ('Framework for state-of-the-art sentence, text, and image embeddings... Provides 5000+ pre-trained models') and there is a real trigger clause ('Use for generating embeddings for RAG, semantic search, or similarity tasks'). It sits between anchors 4 and 5 because the 'when' is task-oriented rather than user-utterance-oriented ('Use for...' vs 'Use when the user mentions...'), so it could be more explicit as trigger guidance.

4 / 5

Trigger Term Quality

Good natural-keyword coverage: 'embeddings', 'RAG', 'semantic search', 'similarity', 'clustering', 'pre-trained models', 'multilingual', 'multimodal' are phrases users actually say. It misses common synonyms like 'vector search', 'text vectors', or 'HuggingFace', so it does not reach the comprehensive-with-synonyms anchor of 5.

4 / 5

Distinctiveness Conflict Risk

It occupies a clear niche (local, pre-trained-model embedding generation) with distinct triggers (RAG, semantic search, similarity). Minor overlap risk remains with competing embedding skills (OpenAI/Instructor/Cohere embeddings) since the description doesn't state its local/no-API differentiator, keeping it just below the minimal-conflict anchor.

4 / 5

Total

16

/

20

Passed

Validation

87%

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

CriteriaDescriptionResult

metadata_version

'metadata.version' is missing

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/llm-tools/sentence-transformers/SKILL.md

Reviewed

Table of Contents

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.