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.

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/data-and-model-poisoning/SKILL.md

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.

The body is an efficient, actionable security skill with executable discovery commands, a ready PoC payload, and an explicit validation contract. Structure is clean but entirely inlined with no progressive split into reference files.

Suggestions

Consider extracting the CVSS table and per-vector exploitation details into a separate reference file (e.g., references/cvss.md) and linking from the body to improve progressive disclosure.

Optionally present the audit workflow as a single numbered sequence with explicit validation checkpoints (e.g., "validate each ingestion point before escalating to PoC") to strengthen workflow clarity.

Dimension	Reasoning	Score
Conciseness	Lean bullets, tables, and code blocks assume Claude's competence without explaining LLM/RLHF basics; a few prose passages (e.g., the supply-chain distinction, chain promotion) could be trimmed slightly.	4 / 5
Actionability	Provides executable grep patterns for ingestion discovery and a copy-paste-ready bash curl PoC loop, with concrete per-ingestion-point audit questions covering common cases.	5 / 5
Workflow Clarity	Clear recognize → audit → exploit → validate flow with an explicit validate_finding contract (success/negative patterns), though presented as sectioned phases rather than a single linear numbered checklist with explicit checkpoints.	4 / 5
Progressive Disclosure	No bundle files exist; content is well-organized into eight numbered sections with clear navigation, but all material is inlined with no separation of bulk reference material.	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.

The description is specific and well-targeted to a distinct security niche, with strong concrete framing and natural trigger terms. Its main weakness is the absence of an explicit "Use when..." clause, which caps completeness.

Suggestions

Add an explicit "Use when..." clause to the description (e.g., "Use when auditing LLM training pipelines, feedback ingestion, or RLHF/continuous-learning loops for poisoning risk") to lift completeness above 3.

Consider folding a few of the metadata when_to_use keywords (e.g., "training data poisoning", "backdoor") into the description prose for broader natural-term coverage.

Dimension	Reasoning	Score
Specificity	Names the domain and several concrete mechanisms ("Hunt LLM training-data and model poisoning", "fine-tuning, RLHF, or continuous-learning loops") with specific attack framing, though not an exhaustive list of distinct actions.	4 / 5
Completeness	The "what" is clear and concrete, but there is no "Use when..." clause or equivalent explicit trigger guidance in the description, capping completeness at 3 per the rubric guidelines.	3 / 5
Trigger Term Quality	Includes natural terms a security tester would say ("model poisoning", "data poisoning", "fine-tuning", "RLHF", "continuous-learning") with good coverage, but misses a few common synonyms like "training data poisoning" phrasing.	4 / 5
Distinctiveness Conflict Risk	Targets a clear niche (OWASP LLM04:2025, LLM training-data and model poisoning) with distinct triggers and minimal overlap risk with 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/data-and-model-poisoning/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.