CtrlK
BlogDocsLog inGet started
Tessl Logo

data-and-model-poisoning

Hunt LLM training-data and model poisoning (OWASP LLM04:2025) — adversarial inputs that bias future model behaviour through fine-tuning, RLHF, or continuous-learning loops.

73

Quality

91%

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

SKILL.md
Quality
Evals
Security

Quality

Content

100%

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A tight, actionable offensive-security playbook: executable audit commands, a runnable PoC, and a validation contract with negative controls, organized into clear numbered sections with no wasted prose.

DimensionReasoningScore

Conciseness

The body is lean — bullet signals, compact code blocks, and a tight intro that distinguishes poisoning from supply-chain compromise without explaining concepts Claude already knows; every section earns its tokens.

3 / 3

Actionability

Provides concrete, copy-paste-ready guidance: executable grep -rE commands in the audit workflow and a runnable curl-loop PoC payload with real flags and JSON body, not pseudocode.

3 / 3

Workflow Clarity

A clear recognition → vectors → audit → PoC → validate sequence, and section 6 defines a validate_finding contract with success_patterns plus negative_command/negative_patterns as explicit validation checkpoints.

3 / 3

Progressive Disclosure

No bundle files exist and none are needed; the skill is a single cohesive file organized into eight clearly numbered sections with no nested/deep references, satisfying the well-organized-sections standard.

3 / 3

Total

12

/

12

Passed

Description

82%

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-targeted description that names concrete poisoning mechanisms and a clear OWASP niche. Its only gap is the absence of an explicit 'Use when…' trigger clause, which caps completeness.

Suggestions

Append an explicit trigger clause such as 'Use when auditing LLM training pipelines, feedback loops, or RAG ingestion for poisoning risk' to lift completeness from 2 to 3.

Consider adding a couple of common phrasings users might say (e.g. 'model backdoor', 'RLHF manipulation') to broaden natural-keyword coverage.

DimensionReasoningScore

Specificity

Names a concrete action ('Hunt LLM training-data and model poisoning') and enumerates multiple specific mechanisms — 'fine-tuning, RLHF, or continuous-learning loops' — matching the 'lists multiple specific concrete actions' anchor.

3 / 3

Completeness

It clearly answers 'what does this do' but lacks any explicit 'Use when…' clause or equivalent trigger guidance, so per the judging guidelines completeness is capped at 2 rather than reaching the explicit-when level of 3.

2 / 3

Trigger Term Quality

Covers natural domain terms a security tester would actually say — 'training-data', 'model poisoning', 'fine-tuning', 'RLHF', 'continuous-learning', 'adversarial inputs' — giving good coverage rather than jargon-only phrasing.

3 / 3

Distinctiveness Conflict Risk

The OWASP LLM04:2025 framing carves a clear niche with distinct triggers, making it unlikely to fire for unrelated security skills; not the level below because it is far more specific than 'works with document files'.

3 / 3

Total

11

/

12

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.

Validation15 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

metadata_version

'metadata.version' is missing

Warning

Total

15

/

16

Passed

Repository
PurpleAILAB/Decepticon
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.