CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

supply-chain

Hunt LLM supply-chain compromise (OWASP LLM03:2025) — malicious or backdoored models, datasets, adapters, plugins, MCP servers, and tokenizer / framework dependencies that ship inside an AI-integrated product.

Quality

78%

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/supply-chain/SKILL.md
SKILL.md
Quality
Evals
Security

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 lean, highly actionable audit skill: concrete inventory commands, executable PoCs, explicit positive/negative validation contracts, and a well-sectioned single-file structure. Only minor issues hold it back — a few undefined placeholders in code examples and slightly overweight inline material (PoC payloads, CVSS tables) that could be split into reference files.

DimensionReasoningScore

Conciseness

The body is dense and assumes competence — recognition signals, attack vectors, audit greps, PoC code, a CVSS table, and a chain-promotion note with no padding or beginner-level explanation. Minor trimming is possible (the opening trust-boundary paragraph and the one-line definition of typo-squatting in 'Typo-squatted framework package' restate things Claude can infer), which keeps it at anchor 4 rather than 5.

4 / 5

Actionability

Concrete, mostly copy-paste-ready guidance throughout: four grep inventory commands, an executable Python trigger-sweep script, a bash typo-squat check, a per-artifact 3-question checklist, indicator/CVSS tables, and a validate_finding contract. Minor gaps — the Python snippet's 'API' variable is undefined, the bash snippet depends on an unspecified 'known-good.txt', and the glob '/workspace/**/*.txt' requires globstar — place it just below anchor 5.

4 / 5

Workflow Clarity

The audit workflow is clearly sequenced (inventory greps by artifact class → per-artifact origin/push-rights/eval checklist → PoC payload → validate) and closes with an explicit validation contract: success_patterns, a negative_command ('same query / build with the suspicious artifact removed'), and negative_patterns that define a feedback loop. This matches anchor 5's checklist + explicit validation + error-recovery pattern; no destructive/batch cap applies since negative controls are present.

5 / 5

Progressive Disclosure

No bundle files exist and the single 145-line body is cleanly sectioned (signals → vectors → workflow → goals → PoCs → validation → CVSS → chaining), with one clearly signaled cross-reference to the prompt-injection skill. It is not anchor 5 because the simple-skill exception is scoped to under-50-line skills: material such as the PoC payloads and the CVSS variant table could live in reference files to slim the main body, i.e. good structure with minor organization gaps.

4 / 5

Total

17

/

20

Passed

Description

66%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 tightly scoped, domain-anchored description with excellent target enumeration and good natural keyword coverage, weakened only by the absence of an explicit 'Use when...' trigger clause in the description text itself. Adding a natural-language trigger clause and a few common synonyms would round it out.

Suggestions

Append an explicit trigger clause to the description, e.g. 'Use when auditing AI products for supply-chain compromise: backdoored or typo-squatted model/adapter/dataset artifacts, unpinned LLM framework dependencies, or unvetted plugin/MCP marketplace installs.'

Include common user-facing synonyms in the description — 'typosquat', 'dependency confusion', 'poisoned dataset', 'HuggingFace model' — so natural phrasings beyond the OWASP term match.

If metadata.when_to_use is intended as trigger guidance, replace the raw keyword dump with the same natural phrases added to the description, so the guidance is consistent and readable.

DimensionReasoningScore

Specificity

The description names the domain ('Hunt LLM supply-chain compromise (OWASP LLM03:2025)') and comprehensively enumerates concrete compromise targets — 'malicious or backdoored models, datasets, adapters, plugins, MCP servers, and tokenizer / framework dependencies that ship inside an AI-integrated product'. It sits above anchor 3 (which expects non-comprehensive coverage) but below anchor 5, since it relies on a single action verb ('Hunt') rather than multiple distinct actions.

4 / 5

Completeness

The 'what' is clear and specific, but the description itself contains no 'Use when...' clause or equivalent explicit trigger guidance — the when is only weakly implied by 'that ship inside an AI-integrated product'. The frontmatter's metadata.when_to_use field is a raw keyword dump outside the description, not natural trigger phrasing, so the cap at 3 applies. It is not a 2 because the 'what' half is concrete and well-scoped, and not a 4 because no explicit 'when' clause exists in the description text.

3 / 5

Trigger Term Quality

Strong natural terms a security reviewer would actually say: 'supply-chain compromise', 'malicious', 'backdoored model', 'dataset', 'adapter', 'plugin', 'MCP servers', 'framework dependencies', plus the 'OWASP LLM03' reference. A few common synonyms are missing (e.g. 'typosquat', 'dependency confusion', 'poisoned dataset', 'HuggingFace'), keeping it below the comprehensive-synonym anchor 5.

4 / 5

Distinctiveness Conflict Risk

The OWASP LLM03 supply-chain framing carves a clear niche with distinct triggers ('backdoored model', 'supply-chain compromise', 'LoRA adapter'), but there is minor overlap risk with closely related skills the body itself acknowledges (e.g. prompt-injection via plugin/MCP tool descriptions, and generic malicious-dependency scanning). This matches anchor 4 rather than 5's 'minimal conflict risk'.

4 / 5

Total

15

/

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/supply-chain/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.