CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

assessing-vector-and-embedding-weaknesses

Test RAG vector stores (Pinecone, Qdrant, Weaviate, Chroma, pgvector, FAISS) for embedding inversion, cross-tenant data leakage, and data poisoning per OWASP LLM08:2025. Use when performing an authorized security assessment of a RAG pipeline's retrieval layer or auditing multi-tenant vector-store isolation.

Quality

90%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Low

Low-risk findings worth noting

SKILL.md
Quality
Evals
Security

Quality

Content

85%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.

The body is highly actionable with executable code and a well-sequenced, validated workflow, but it underuses its own bundle files, inlining standards/API detail that already lives in references/ without linking to them.

Suggestions

Replace the inlined MITRE ATT&CK Mapping table and Tools-and-Resources API detail with a one-line pointer to references/standards.md and references/api-reference.md so the body stays an overview.

Add a "References" section linking references/api-reference.md, references/standards.md, and scripts/agent.py (and noting what each covers) so the bundle is discoverable from SKILL.md.

Tighten the Overview's attack-class bullets to drop definitional phrasing Claude already knows (e.g. "embeddings are not one-way") and keep only the assessment-relevant framing.

DimensionReasoningScore

Conciseness

Mostly efficient with concrete code throughout, but the Overview explains the five attack classes with some phrasing that could be trimmed (e.g. "embeddings are not one-way"), sitting just below the lean anchor.

4 / 5

Actionability

Each of the seven steps ships copy-paste-ready, executable Python (qdrant-client, cosine_similarity, Filter/FieldCondition, PointStruct upsert, regex injection scan) covering the common cases.

5 / 5

Workflow Clarity

A clear seven-step sequence with explicit validation checkpoints (Step 4's unfiltered re-test to confirm server-side isolation, "test collection only" guard, and the final Validation Criteria checklist) and feedback for the destructive/batch operations.

5 / 5

Progressive Disclosure

The body is well sectioned but never signals the existing bundle files (references/api-reference.md, references/standards.md, scripts/agent.py); the MITRE/standards mapping and API patterns are inlined and duplicated in the bundle with no navigation, fitting the "references not clearly signaled / content that should be separate is inline" anchor.

3 / 5

Total

17

/

20

Passed

Description

92%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, specific description that names the target systems, the three weakness classes, the governing standard, and explicit use-when triggers in third-person voice. The only mild gap is keyword synonym coverage.

DimensionReasoningScore

Specificity

"Test RAG vector stores (Pinecone, Qdrant, Weaviate, Chroma, pgvector, FAISS) for embedding inversion, cross-tenant data leakage, and data poisoning" lists multiple concrete test actions across six named stores, matching the comprehensive-coverage anchor.

5 / 5

Completeness

It states a clear "what" (test the three weakness classes per OWASP LLM08:2025) and an explicit "Use when..." trigger clause with concrete scenarios, matching the both-what-and-when anchor.

5 / 5

Trigger Term Quality

Natural terms for the security audience appear ("authorized security assessment of a RAG pipeline's retrieval layer", "auditing multi-tenant vector-store isolation") with named stores standing in for file-extension equivalents, but a few common phrasings/synonyms are missing so it sits just below the comprehensive anchor.

4 / 5

Distinctiveness Conflict Risk

The RAG vector-store / OWASP LLM08 / embedding-inversion niche is narrow with distinct triggers and minimal overlap with other skills.

5 / 5

Total

19

/

20

Passed

Validation

93%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 15 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

frontmatter_unknown_keys

Unknown frontmatter key(s) found; consider removing or moving to metadata

Warning

Total

15

/

16

Passed

Repository

mukul975/Anthropic-Cybersecurity-Skills

Path

skills/assessing-vector-and-embedding-weaknesses/SKILL.md

Revision

54a7988

Reviewed

Table of Contents

DiscoveryImplementationValidation

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.

Claim skill