CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

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

80%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

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
SKILL.md
Quality
Evals
Security

Quality

Content

82%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 lean, highly actionable playbook: concrete greps, ready PoC payloads, explicit validation/negative-control contract, and CVSS scoring, with clear section organization and no bundle-file overhead. The only refinement opportunities are trimming minor rhetorical color in the intro and possibly splitting the payload library and CVSS defaults into reference files.

DimensionReasoningScore

Conciseness

The body is dense and assumes competence — no space is spent explaining RAG, CVSS, or embeddings; content is delivered as signal lists, grep commands, tables, and payloads. A few rhetorical flourishes ('the model is one creative prompt away from emitting them verbatim', 'the model dutifully repeats the result') and the two-sentence prioritization intro could be trimmed slightly, matching the 'efficient; minor instances of over-explanation' anchor rather than the score-5 every-token-earns-its-place anchor.

4 / 5

Actionability

Everything is copy-paste ready: four executable grep pipelines for finding retrieval call sites, tenancy filters, and debug echoes; ready-to-send PoC payloads (system-prompt extraction, cross-tenant marker probe, memorisation oracle); concrete success patterns ('sk-[A-Za-z0-9]{20,}', 'BEGIN (PRIVATE|RSA)') and a full CVSS table. This matches the 'fully executable; copy-paste ready; covers common cases' anchor.

5 / 5

Workflow Clarity

Sections are numbered 1–8 in a sensible engagement order (recognise → vectors → audit → exploit → validate → score → chain), and validation is explicit: the validate_finding contract defines negative_command/negative_patterns, and the cross-tenant probe includes a two-run confirmation loop. It stops short of score 5 because the end-to-end sequence is implied by section numbering rather than framed as an explicit ordered procedure, and the validation contract is not wired into a 'fix and re-run' loop — minor gaps matching the score-4 anchor.

4 / 5

Progressive Disclosure

No bundle files (references/, scripts/, assets/) exist and the body references none, so the skill is self-contained; the eight clearly headed sections make navigation easy and the ~130-line playbook is reasonably placed inline. It is not score 5 because some quasi-reference material (the PoC payload library and the default CVSS table) is inlined where a sibling-skills convention of separate payload/CVSS references would fit — 'good structure; most content appropriately placed; minor organization gaps'.

4 / 5

Total

17

/

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.

A specific, well-scoped description with excellent distinctiveness and good trigger vocabulary, weakened only by the absence of an explicit 'Use when...' clause in the description field itself (trigger guidance lives in metadata.when_to_use instead). Adding a when-to-use sentence with common synonyms would lift it to the top anchor.

Suggestions

Append an explicit trigger clause to the description, e.g. 'Use when auditing LLM-backed products for data leaks, prompt/system-prompt disclosure, cross-tenant retrieval leaks, or training-data memorisation.'

Add natural trigger synonyms users are likely to say — 'data leak', 'prompt leak', 'system prompt leak', 'cross-tenant', 'RAG leak' — to improve trigger term coverage.

Mention system-prompt disclosure explicitly in the what-list, since it is one of the most commonly tested variants and currently only appears in the body.

DimensionReasoningScore

Specificity

The description names one concrete action ('Hunt LLM sensitive-information disclosure (OWASP LLM02:2025)') and enumerates the specific leakage targets ('PII, secrets, internal source, model details, and other-tenant data') and mechanisms ('model outputs, training-data extraction, or retrieval-side joins'). Coverage of the vulnerability class is strong with only minor gaps (e.g., no mention of system-prompt leaks or deliverables), matching the 'several specific actions; minor gaps' anchor rather than the comprehensive score-5 anchor and well above the generic score-3 anchor.

4 / 5

Completeness

The 'what' is clear and concrete (hunting this specific OWASP disclosure class with enumerated data types and mechanisms), but the description field itself contains no 'Use when...' clause or equivalent explicit trigger guidance — per the judging guidelines this caps completeness at 3. It is not score 2 because the 'what' is specific rather than vague; not score 4 because no 'when' is stated in the description itself.

3 / 5

Trigger Term Quality

It contains natural terms users would say: 'LLM', 'PII', 'secrets', 'leakage', 'training data', 'other-tenant', 'retrieval', plus the OWASP LLM02:2025 identifier. A few common variations are missing ('data leak', 'prompt leak', 'system prompt leak', 'cross-tenant'), which fits the 'good keyword coverage; a few natural terms missing' anchor rather than the score-5 comprehensive-synonyms anchor.

4 / 5

Distinctiveness Conflict Risk

The description pins a clear niche — OWASP LLM02:2025 sensitive-information disclosure — with distinct trigger vocabulary (PII, training-data extraction, other-tenant, retrieval-side joins) that does not overlap with sibling categories like prompt injection (LLM01) or generic security skills. This matches the 'clear niche with distinct triggers; minimal conflict risk' anchor.

5 / 5

Total

16

/

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

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

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.