vector-and-embedding-weaknesses

Hunt vector / embedding weaknesses (OWASP LLM08:2025) — adversarial inputs against the RAG / similarity layer that cause cross-tenant leak, embedding-inversion privacy loss, semantic confusion, and retriever-driven prompt injection.

Quality

76%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Critical

Do not install without reviewing

Fix and improve this skill with Tessl

tessl review fix ./packages/decepticon/decepticon/skills/standard/analyst/vector-and-embedding-weaknesses/SKILL.md

Quality

Content

81%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 tightly structured, actionable hunting playbook with strong workflow clarity and built-in finding validation, held back only by minor framing verbosity and placeholder gaps in the PoC code.

Suggestions

Tighten the opening paragraph and section 8 prose to remove motivational framing that does not directly guide action.

Replace placeholders in the embedding-inversion PoC (LEAKED_EMB, [...candidate strings...]) with a minimal concrete example so the snippet is copy-paste runnable.

Consider extracting the CVSS table and/or PoC payloads into a reference file to improve progressive disclosure for this multi-faceted skill.

Dimension	Reasoning	Score
Conciseness	Dense and information-rich with executable grep commands, PoC payloads, and a CVSS table, but the opening motivational paragraph and section 8 editorial framing add minor trimmable overhead.	4 / 5
Actionability	Provides copy-paste-ready grep recon commands and concrete audit checklists, but the PoC Python snippet uses placeholders (LEAKED_EMB, [...candidate strings...]) leaving minor gaps.	4 / 5
Workflow Clarity	Clear recon → per-retrieval/per-indexer checklists → exploitation → validation sequence, with an explicit validate_finding contract (negative_command, negative_patterns) serving as a validation feedback loop.	5 / 5
Progressive Disclosure	Well-organized into eight numbered sections with no nested references and no bundle files; content is self-contained, though CVSS/PoC detail could optionally be split into reference files.	4 / 5
	Total	17 / 20 Passed

Description

71%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 specific, well-scoped security hunting description with strong concrete coverage of the LLM08 attack surface, weakened only by the absence of an explicit "Use when" trigger clause in the description field.

Suggestions

Add an explicit "Use when..." clause to the description (e.g. "Use when auditing RAG/vector-store security, OWASP LLM08, or cross-tenant retrieval and embedding-inversion risks") to lift completeness above 3.

Include a couple more natural trigger phrases such as "vector database", "RAG poisoning", and "embedding attack" to broaden trigger-term coverage.

Sharpen distinctiveness by scoping the prompt-injection reference to its index-time/vector form to reduce overlap with general LLM01 prompt-injection skills.

Dimension	Reasoning	Score
Specificity	Names a concrete action ("Hunt vector / embedding weaknesses") and enumerates four specific concrete outcomes — cross-tenant leak, embedding-inversion privacy loss, semantic confusion, and retriever-driven prompt injection — giving comprehensive coverage of the domain.	5 / 5
Completeness	The description gives a clear, concrete "what" but lacks any "Use when..." clause or explicit trigger guidance in the description field itself (trigger terms live only in the metadata when_to_use), capping completeness at 3.	3 / 5
Trigger Term Quality	Good keyword coverage across vector, embedding, RAG, similarity, cross-tenant leak, inversion, retriever, and prompt injection, with a few natural variations missing (e.g. "vector database", "RAG poisoning").	4 / 5
Distinctiveness Conflict Risk	It carves a clear OWASP LLM08 niche around the vector/similarity layer, but explicitly references retriever-driven prompt injection and cross-tenant leak, creating minor overlap with LLM01/LLM02 skills.	4 / 5
	Total	16 / 20 Passed

Validation

93%

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 — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
metadata_version	'metadata.version' is missing	Warning

	Total	15 / 16 Passed

Repository: PurpleAILAB/Decepticon
Path: packages/decepticon/decepticon/skills/standard/analyst/vector-and-embedding-weaknesses/SKILL.md
Commit: 0cf691e

Reviewed: 2 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.