sensitive-information-disclosure

Hunt LLM sensitive-information disclosure (OWASP LLM02:2025) — leakage of PII, secrets, internal source, model details, and other-tenant data through model outputs, training-data extraction, or retrieval-side joins.

Quality

72%

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/sensitive-information-disclosure/SKILL.md

Quality

Content

75%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 content is a lean, highly actionable security-testing runbook with concrete commands, payloads, CVSS vectors, and an explicit validation contract. Its main limitation is that two attack vectors remain prose-only and the workflow is sectioned by topic rather than presented as a strict numbered pipeline with per-stage checkpoints.

Suggestions

Add executable commands or one-liners for the embedding-inversion and error-oracle vectors so every attack vector has runnable guidance, not just prose.

Reformat the audit workflow as a single numbered end-to-end pipeline with an explicit 'Validate the finding before promoting' checkpoint between exploitation and chain promotion.

Consider externalising the PoC payloads and CVSS tables into a references/ file so SKILL.md stays a lean overview with one-level-deep navigation.

Dimension	Reasoning	Score
Conciseness	The body is dense and assumes Claude's security competence, with copy-paste grep blocks and compact payloads; only the opening framing paragraph and a couple of editorial lines ('Disclosure usually beats prompt injection...') could be trimmed.	4 / 5
Actionability	Executable grep commands, copy-paste PoC prompt strings, and concrete regex success/negative patterns cover the common cases, but the embedding-inversion and error-oracle vectors are described in prose without runnable commands, leaving minor gaps.	4 / 5
Workflow Clarity	A clear recognize → choose-vector → exploit → validate sequence is present, with the 'validate_finding' contract (success_patterns, negative_command, negative_patterns) acting as the validation checkpoint; the sequence is sectioned by topic rather than a strictly numbered end-to-end pipeline, leaving minor gaps.	4 / 5
Progressive Disclosure	A single self-contained SKILL.md is organised into eight clearly numbered, navigable sections with no nested references; at ~127 lines it is denser than the 'simple skill' case, so it sits just below the well-organised-sections ceiling of 5.	4 / 5
	Total	16 / 20 Passed

Description

70%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.

The description is specific, well-targeted to a distinct OWASP niche, and rich in relevant trigger terms, but it omits an explicit 'Use when...' trigger clause, which caps completeness. Adding a concrete usage-trigger sentence would lift the weakest dimension.

Suggestions

Append an explicit 'Use when...' clause naming concrete trigger phrases, e.g. 'Use when testing LLM apps for data leakage, cross-tenant RAG disclosure, system-prompt extraction, or training-data memorisation.'

Add natural-synonym trigger terms ('data leak', 'data exposure', 'prompt leakage') alongside the technical jargon to broaden when-to-use matching.

Consider a second action verb (e.g. 'Hunt and validate...') to move specificity from one enumerated action toward multiple concrete actions.

Dimension	Reasoning	Score
Specificity	A single concrete verb ('Hunt') is used, but it enumerates five specific disclosure targets (PII, secrets, internal source, model details, other-tenant data) across three mechanisms, giving broad coverage with only a minor gap in action-verb diversity.	4 / 5
Completeness	The 'what' is clearly stated ('Hunt LLM sensitive-information disclosure...') but there is no 'Use when...' clause or equivalent explicit trigger guidance, capping completeness at 3 per the rubric guideline.	3 / 5
Trigger Term Quality	Strong security-niche keywords ('sensitive-information disclosure', 'PII', 'secrets', 'training-data extraction') that a tester would naturally use, though natural synonyms like 'leak' or 'exposure' are absent.	4 / 5
Distinctiveness Conflict Risk	The OWASP LLM02:2025 reference pins a clear, narrow niche (disclosure vs. injection) with distinct triggers and minimal overlap risk against other skills.	5 / 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/sensitive-information-disclosure/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.