moai-ref-llm-security

AI/LLM defensive security reference: prompt-injection defense, OWASP LLM Top 10 defensive mapping, MCP and agentic tool-call hardening, training-data poisoning detection, model-output validation and guardrails, MITRE ATLAS defensive correlation, and NIST AI RMF governance. Agent-extending skill that amplifies backend, security, and AI-application engineering with production-grade defensive patterns for LLM-backed systems. NOT for: offensive techniques (jailbreak authoring, attack-payload crafting, red-team exploitation), model training or fine-tuning methodology, prompt optimization for capability, web-app OWASP Top 10 (see moai-ref-owasp-checklist), or general API design (see moai-ref-api-patterns).

Quality

85%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Quality

Content

71%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, defensive-only reference that is dense and actionable with strong validation framing (checklist, red flags, severity levels). Its main weakness is token efficiency: the untrusted-by-default thesis is repeated across sections and no detail is offloaded to bundle files despite progressive disclosure being enabled.

Suggestions

Split the large OWASP LLM Top 10, MITRE ATLAS, and NIST AI RMF mapping tables into separate reference files under references/ and link to them from the body, realizing the declared progressive_disclosure levels and trimming the inline token cost.

Consolidate the repeated 'treat every text channel as untrusted data' framing into a single statement in the Trust Boundaries section and reference it from later sections instead of restating it.

Consider condensing the opening 'Target Use' section, which overlaps with the description's when_to_use framing and the Trust Boundaries threat-model paragraph.

Dimension	Reasoning	Score
Conciseness	Dense, practitioner-focused tables and tight prose with no jargon-padding or explaining of basics Claude already knows; docked one point because the 'treat every channel as untrusted data' thesis is restated across the intro, Trust Boundaries, Prompt-Injection, and closing sections more than necessary.	4 / 5
Actionability	As an instruction-only reference it provides concrete, named controls per risk (instruction-hierarchy enforcement, provenance tagging, schema-validate tool results, least-privilege tool set, canary artifacts, context-aware encoding) tied to specific OWASP/ATLAS IDs, rather than abstract direction — actionable though without copy-paste code.	4 / 5
Workflow Clarity	An explicit Verification checklist, Red Flags list, and Defensive Severity Levels give clear validation checkpoints for a hardening review, but there is no strict numbered validate->fix->retry sequence, so it sits just below the top anchor.	4 / 5
Progressive Disclosure	Sections are well-organized with clear headers and there are no nested references, but the ~250-line body keeps all detail inline with no bundle files in references/scripts/assets, so content that could be split into separate reference files (e.g. the OWASP/ATLAS/NIST tables) is not offloaded despite progressive_disclosure being declared enabled.	3 / 5
	Total	15 / 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.

The description is exemplary: third-person, concrete, comprehensive, and explicitly distinguishes both what it covers and when to invoke it, with sibling-skill disambiguation. It would naturally surface for the right LLM-hardening tasks and avoid conflicts.

Dimension	Reasoning	Score
Specificity	Lists seven concrete defensive capabilities (prompt-injection defense, OWASP LLM Top 10 mapping, MCP/agentic tool-call hardening, training-data poisoning detection, output validation/guardrails, MITRE ATLAS correlation, NIST AI RMF governance) plus an explicit 'NOT for' exclusion list — comprehensive concrete actions, not vague.	5 / 5
Completeness	Explicitly answers both 'what' (defensive security reference across seven named areas) and 'when' ('Use when hardening... designing guardrails... scoping MCP/tool permissions...'), with concrete trigger phrases in both directions.	5 / 5
Trigger Term Quality	The 'when_to_use' clause surfaces exactly the natural phrases a user would say — 'hardening an LLM-backed application', 'prompt injection', 'guardrails', 'output validation', 'MCP/tool permissions', 'OWASP LLM Top 10', 'NIST AI RMF', 'MITRE ATLAS' — with strong synonym and acronym coverage.	5 / 5
Distinctiveness Conflict Risk	Clear LLM-defensive-security niche with explicit disambiguation from sibling skills (moai-ref-owasp-checklist, moai-ref-api-patterns) and an exclusion of offensive techniques, minimizing wrong-skill triggering.	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
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: modu-ai/moai-adk
Path: .claude/skills/moai-ref-llm-security/SKILL.md
Commit: 4100d87

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.