adversarial-ml-evasion

Craft adversarial examples that cause trained ML classifiers to misclassify at inference time — image recognition, malware detectors, IDS, spam filters.

Quality

74%

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/adversarial-ml-evasion/SKILL.md

Quality

Content

78%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 highly actionable with executable examples and a clear sequenced workflow plus validation checklist, but it is a monolithic single file with no progressive disclosure into reference bundles.

Suggestions

Extract bulk reference material (the tooling table and detection signals section) into separate reference files under references/ and link to them, improving progressive disclosure.

Add an explicit validate→fix→retry feedback loop template that applies consistently across attack types, not just the malware perturbation section.

Tighten a few explanatory inline comments (e.g., 'Confidence scores = white-box-equivalent gradient signal') to push conciseness toward fully lean.

Dimension	Reasoning	Score
Conciseness	The body is efficient and avoids explaining basic concepts Claude already knows; a few inline comments are slightly explanatory but most earn their place, leaving only minor trimmable instances.	4 / 5
Actionability	Provides fully executable, copy-paste-ready code and commands across the common cases (FGSM, PGD, C&W via ART, NES gradient estimation, transfer attacks, HopSkipJump, adversarial patches, malware perturbation, IDS evasion).	5 / 5
Workflow Clarity	A clear numbered sequence (Recon → White/Black-box → Physical → Tooling → Validation → Detection) with an explicit validation checklist is present, but the validate→fix→retry feedback loop is only fully spelled out in one section rather than consistently across attack types.	4 / 5
Progressive Disclosure	The skill is a well-sectioned single file with no bundle references; since it exceeds the under-50-line simple-skill exception and inlines content (tooling table, detection signals) that could live in separate files, structure is good but organization is only mid-level.	3 / 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 and clearly distinguishes the skill's niche, but it omits an explicit 'Use when...' trigger clause, capping completeness at the midpoint.

Suggestions

Add an explicit 'Use when...' clause naming the trigger phrases a user would naturally say (e.g., 'Use when evading ML-based classifiers, AV/EDR/IDS, or generating adversarial examples').

Broaden trigger-term coverage with synonyms users commonly say (e.g., 'fool', 'bypass', 'evade', 'classifier bypass', 'perturbation').

List a couple more concrete actions beyond crafting examples (e.g., estimate gradients black-box, evaluate transfer success) to push specificity toward a comprehensive level.

Dimension	Reasoning	Score
Specificity	Names the domain and a concrete action ('Craft adversarial examples that cause trained ML classifiers to misclassify') plus four specific target domains (image recognition, malware detectors, IDS, spam filters), but the core action is a single one applied across targets rather than a comprehensive list of distinct actions.	4 / 5
Completeness	Provides a clear 'what' but no explicit 'Use when...' clause in the description; the 'when' is only weakly implied through the listed target domains, which per guidelines caps completeness at 3.	3 / 5
Trigger Term Quality	Includes natural, recognizable terms a user would say ('adversarial examples', 'misclassify', 'image recognition', 'malware detectors', 'IDS', 'spam filters') with good coverage, though a few common synonyms are absent.	4 / 5
Distinctiveness Conflict Risk	Occupies a clear niche (adversarial-example evasion of trained classifiers) 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/adversarial-ml-evasion/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.