ctf-ai-ml

Provides AI and machine learning techniques for CTF challenges. Use when attacking ML models, crafting adversarial examples, performing model extraction, prompt injection, membership inference, training data poisoning, fine-tuning manipulation, neural network analysis, LoRA adapter exploitation, LLM jailbreaking, or solving AI-related puzzles.

Quality

82%

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

Quality

Content

72%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, well-organized reference catalog with executable triage commands and clear file-pointing structure, but it is undermined by missing bundle files and lacks explicit validation checkpoints in its workflows. The progressive-disclosure design is sound in form but broken in substance because the referenced detail files are absent.

Suggestions

Add the three referenced supporting files (model-attacks.md, adversarial-ml.md, llm-attacks.md) to the bundle so the well-signaled one-level-deep references resolve instead of dangling.

Insert explicit validation checkpoints in the Quick Start triage (e.g., confirm the model loaded and printed keys before running the weight-comparison step, and verify the remote endpoint responded before interpreting prompt-injection output).

Include at least one complete, copy-paste-ready end-to-end attack snippet inline (e.g., a full FGSM/PGD loop) rather than only the formula plus a link, to lift actionability on the most common attack cases.

Dimension	Reasoning	Score
Conciseness	The body is lean — it gives formulas directly ('x_adv = x + eps * sign(grad_x(loss))', 'W_base + alpha * (B @ A)'), copy-paste commands, and one-line technique summaries with file links, without explaining concepts Claude already knows; it fits the 'lean and efficient; every token earns its place' anchor rather than the 4 anchor's 'minor over-explanation'.	5 / 5
Actionability	The Quick Start section is fully executable (torch/safetensors/transformers inspection, weight comparison, curl prompt-injection probe), but the headline attack techniques are presented only as formulas/descriptions with links to detail files rather than complete inline implementations, fitting 'mostly executable guidance; minor gaps' above the pseudocode 3 anchor.	4 / 5
Workflow Clarity	The Quick Start offers an ordered triage sequence and 'When to Pivot' provides routing, but there are no explicit validation/verification checkpoints or feedback loops for the attack processes, matching the 3 anchor ('sequence present but checkpoints missing or implicit') rather than the 4 anchor which requires most checkpoints present.	3 / 5
Progressive Disclosure	The structure is well-signaled with an 'Additional Resources' index and one-level-deep links to model-attacks.md, adversarial-ml.md, and llm-attacks.md, but those supporting files are not present in the bundle (no references/ dir, no other .md files exist), so the references dangle and the overview→detail promise is unfulfilled — below the 4 anchor whose 'minor gaps' assume the referenced files actually exist.	3 / 5
	Total	15 / 20 Passed

Description

92%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 strong, specific description that clearly states what the skill does and when to use it, with comprehensive concrete technique triggers and a well-scoped niche. Its only weakness is trigger-term phrasing that favors technical jargon over the casual synonyms a less expert user might say.

Dimension	Reasoning	Score
Specificity	Lists many concrete actions ('attacking ML models, crafting adversarial examples, performing model extraction, prompt injection, membership inference, training data poisoning, fine-tuning manipulation, neural network analysis, LoRA adapter exploitation, LLM jailbreaking') — comprehensive coverage of the AI/ML CTF domain, matching the 'multiple specific concrete actions' anchor rather than the 4 anchor which expects minor gaps.	5 / 5
Completeness	Explicitly answers both 'what' ('Provides AI and machine learning techniques for CTF challenges') and 'when' ('Use when attacking ML models...') with concrete trigger phrases, matching the 5 anchor exactly rather than the 4 anchor where 'when' is only loosely specified.	5 / 5
Trigger Term Quality	Strong domain-specific trigger coverage (prompt injection, adversarial examples, model extraction, LLM jailbreaking, membership inference) that the target CTF audience would naturally say, but it leans technical and omits casual synonyms a user might phrase (e.g. 'AI challenge', 'ML hacking'), so it sits above the 3 anchor but short of the comprehensive-synonym 5 anchor.	4 / 5
Distinctiveness Conflict Risk	Scoped to a clear niche (AI/ML techniques for CTF challenges) with distinct triggers and minimal overlap with sibling skills; the body's 'When to Pivot' section further separates it from /ctf-crypto, /ctf-reverse, and /ctf-misc, fitting the 'clear niche with distinct triggers' anchor.	5 / 5
	Total	19 / 20 Passed

Validation

87%

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 — 14 / 16 Passed

Validation for skill structure

Criteria	Description	Result
metadata_version	'metadata.version' is missing	Warning
relative_links	Relative link issues: 19 missing	Warning

	Total	14 / 16 Passed

Repository: ljagiello/ctf-skills
Path: ctf-ai-ml/SKILL.md
Commit: 36c72e5

Reviewed: 4 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.