ai-security

Use when assessing AI/ML systems for prompt injection, jailbreak vulnerabilities, model inversion risk, data poisoning exposure, or agent tool abuse. Covers MITRE ATLAS technique mapping, injection signature detection, and adversarial robustness scoring.

Quality

87%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

—

The risk profile of this skill

Quality

Content

75%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 well-structured, highly actionable skill body with executable scanner commands, clear workflows, and proper progressive disclosure to a real bundle reference. Its main weakness is conciseness — several concept-explainer paragraphs restate AI-security fundamentals Claude already knows.

Suggestions

Trim the definitional prose in Prompt Injection Detection, Model Inversion Risk, and Data Poisoning Risk (e.g. 'Prompt injection occurs when...', 'Model inversion attacks reconstruct...') to one-line leads; Claude already knows these concepts.

Add an explicit validate->fix->retry loop to Workflow 1/Phase 1 (e.g. on exit code 1 or 2, remediate findings and re-run the scanner until clean) to satisfy the feedback-loop expectation for batch/destructive assessment.

Move the full Injection Signature Categories and Jailbreak Taxonomy tables into a reference file (like atlas-coverage.md) and keep only the handful of signatures needed for quick use inline, tightening token cost.

Dimension	Reasoning	Score
Conciseness	The body is lengthy and several prose sections restate domain knowledge Claude already has (e.g. explaining what prompt injection is, what model inversion attacks reconstruct, what data poisoning does), which is padding that could be trimmed; the tables and code blocks are efficient but the explanatory paragraphs are not.	3 / 5
Actionability	Provides copy-paste-ready scanner invocations with concrete flags, a defined test-file format, exit-code semantics, and jq pipelines wired into CI/CD gating — fully executable guidance covering common cases.	5 / 5
Workflow Clarity	Workflows are clearly sequenced with explicit decision/branching guidance (exit-code decisions, CI/CD gating) and authorization prerequisites, but the static-analysis workflow lacks an explicit validate-then-fix retry loop for findings, leaving a minor validation gap relative to the rubric's feedback-loop anchor.	4 / 5
Progressive Disclosure	Structure is good: a clear overview, inlined quick-start commands, and a well-signaled one-level reference to references/atlas-coverage.md (which exists) for bulk ATLAS detail; minor gaps because some reference-grade tables (full signature/taxonomy catalogs) are inlined rather than split out.	4 / 5
	Total	16 / 20 Passed

Description

100%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 concise, third-person description that pairs an explicit 'Use when' trigger with a comprehensive list of concrete capabilities and the MITRE ATLAS framing. It cleanly answers what the skill does and when to invoke it, with minimal conflict risk.

Dimension	Reasoning	Score
Specificity	Lists multiple specific concrete actions — assessing for prompt injection, jailbreak vulnerabilities, model inversion risk, data poisoning exposure, and agent tool abuse — plus concrete capabilities (ATLAS technique mapping, injection signature detection, adversarial robustness scoring).	5 / 5
Completeness	Opens with an explicit 'Use when assessing...' trigger clause and pairs it with concrete coverage of what the skill does, cleanly answering both 'what' and 'when'.	5 / 5
Trigger Term Quality	Comprehensive natural terms including 'prompt injection', 'jailbreak vulnerabilities', 'model inversion risk', 'data poisoning exposure', 'agent tool abuse', and the recognizable standard 'MITRE ATLAS' that practitioners would name directly.	5 / 5
Distinctiveness Conflict Risk	Occupies a clear niche (AI/ML system and LLM-agent security) with distinct triggers and ATLAS-specific framing; low overlap risk with general app-security or threat-detection skills.	5 / 5
	Total	20 / 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
relative_links	Relative link issues: 4 suspicious	Warning

	Total	15 / 16 Passed

Repository: alirezarezvani/claude-skills
Path: engineering-team/skills/ai-security/SKILL.md
Commit: 19392f7

Reviewed: ~24 hours 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.