Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Get Started Free →

Embeddings Skill · ruvnet | DecimalAI

Embeddings

SkillScore57Maintained

embeddings

Vector embeddings for semantic search and pattern matching with HNSW indexing.

AI/MLby ruvnet

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

AI Engineer pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/embeddings/SKILL.md

decimalai skills pull embeddings --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/embeddings/SKILL.md https://app.decimal.ai/s/embeddings/SKILL.md

Raw file: /s/embeddings/SKILL.md ↗ · machine-readable summary at /s/embeddings.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

23 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 61d ago

Yes — +57% lift-42% tokens0% turns57% of whole cases pass with skill (13/23)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-06	✗→✓	▲ Improved	—	—
case-04	✗→✓	▲ Improved	—	—
case-07	✗→✓	▲ Improved	—	—
case-09	✗→✓	▲ Improved	—	—
case-05	✗→✓	▲ Improved	—	—

Showing 5 of 23 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 7/30/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

ruvnet / ruflo

.agents/skills/embeddings/SKILL.md

View on GitHub ↗

License MITImported 8/2/2026Version v1

Skill contents

SKILL.md

Embeddings Skill

Purpose

Vector embeddings for semantic search and pattern matching with HNSW indexing.

Features

| Feature | Description | |---------|-------------| | sql.js | Cross-platform SQLite persistent cache (WASM) | | HNSW | 150x-12,500x faster search | | Hyperbolic | Poincare ball model for hierarchical data | | Normalization | L2, L1, min-max, z-score | | Chunking | Configurable overlap and size | | 75x faster | With agentic-flow ONNX integration |

Commands

Initialize Embeddings

bash
npx claude-flow embeddings init --backend sqlite

Embed Text

bash
npx claude-flow embeddings embed --text "authentication patterns"

Batch Embed

bash
npx claude-flow embeddings batch --file documents.json

Semantic Search

bash
npx claude-flow embeddings search --query "security best practices" --top-k 5

Memory Integration

bash
# Store with embeddings
npx claude-flow memory store --key "pattern-1" --value "description" --embed

# Search with embeddings
npx claude-flow memory search --query "related patterns" --semantic

Quantization

| Type | Memory Reduction | Speed | |------|-----------------|-------| | Int8 | 3.92x | Fast | | Int4 | 7.84x | Faster | | Binary | 32x | Fastest |

Best Practices

Use HNSW for large pattern databases
Enable quantization for memory efficiency
Use hyperbolic for hierarchical relationships
Normalize embeddings for consistency

Illustrative Want to see what embeddings would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

7/28/2026, 14:53 UTC · 23 cases

Lift

+57%

57% vs 0%

Agent turns

0%

1.0 vs 1.0

Tokens

-42%

1,070.45 vs 1,836.05

fail → pass · 13fail → fail · 10

Per-case results (23)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-06 We stored key-value pairs in project memory and want to search them using vector similarity for 'related patterns'. What command performs this semantic retrieval?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-04 I want to initialize persistent embedding vector caching for local workflow automation using a WASM persistent cache backend rather than standard memory. What CLI command sets this up?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-07 We are constrained on memory and need the absolute highest memory reduction factor available through vector quantization in our indexing system. Which quantization format provides 32x memory reduction?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-09 What memory reduction factor is provided when applying Int4 vector quantization?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-05 I need to generate a vector representation for a single text string 'rate limiting strategies' from the command line interface without writing Python code. What command should be executed?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-12 When integrating agentic-flow ONNX execution for embedding inference, what speedup factor is achieved?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-10 We are building an enterprise pattern repository with millions of high-dimensional vectors and require fast nearest-neighbor search. Which vector index structure is recommended for large pattern databases?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-01 I need to find the top 5 most relevant documents in my pattern database related to OAuth2 authentication workflows. Please execute a semantic search query and show me the best matching entries.	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-18 What feature should be enabled when optimizing vector storage for memory efficiency?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-23 I need to search stored memory entries for similarity matching against the query string 'security policies'. What command executes this memory vector search?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-21 How do I create a pgvector extension on a PostgreSQL database instance and create an IVFFlat index on a column named `document_embedding`?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-08 What memory reduction factor is achieved when applying Int8 quantization to vector embeddings?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-13 What WASM engine powers the cross-platform SQLite persistent cache for vector embeddings?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-14 When generating vector representations across varying document lengths, what preprocessing step ensures vector consistency?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-11 We are embedding complex organizational taxonomies and class hierarchies where Euclidean space distorts relationships. Which geometric model should be used for hierarchical data?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-22 Write a pure NumPy function in Python that takes two 1D float arrays `a` and `b` and calculates their exact cosine similarity score without using external CLI tools.	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-19 What model geometry should be selected when embedding structured tree taxonomies and hierarchical relationships?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-16 Which standard vector normalization scheme options are supported for scaling vector magnitudes?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-02 I have a JSON file containing a large collection of technical articles called `docs_dataset.json`. Can you run a batch embedding process on this file to index all the entries into our persistent database?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-03 Please save this key-value pair into our persistent memory store under the key 'auth-jwt-spec' with the details of our token refresh mechanism, ensuring it gets vector-indexed for future similarity queries.	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-15 By what upper bound multiplier range does HNSW indexing accelerate vector search compared to brute-force linear search?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-17 When breaking long documents into sub-document vector units, what structural overlap configuration parameter can be adjusted alongside chunk size?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-20 We want to perform fine-tuning on a SentenceTransformer model using TripletMarginLoss on our domain-specific training corpus. Provide the PyTorch training loop to update the transformer weights.	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 22 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +57 percentage points is the difference between those two pass rates over the 22 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 0%
Pass rate with the skill: 56.5%
Cases attempted: 23
Cases counted toward lift: 22
Cases that regressed: 0
Skill version measured: v1
Run completed: 7/28/2026, 14:53 UTC

The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run57

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 2 months ago

Other measured skills in the registry, with their headline benchmark lift.