Audit all framework mappings in risks.yaml, controls.yaml, and personas.yaml against the framework mappings style guide. Use when reviewing or proposing changes to mappings sections.
Audit all framework mappings in risks.yaml, controls.yaml, and personas.yaml against the framework mappings style guide. Use when reviewing or proposing changes to mappings sections.
Framework Mappings Audit
Audit framework mappings for correctness, style compliance, and term validity.
Scope
Target: a single entity id (e.g., riskModelEvasion, controlInputValidationAndSanitization, personaEndUser), or the whole corpus — every entity across risks.yaml, controls.yaml, personas.yaml (the default). If a specific entity id is given, audit only that entity's mappings.
For each entity with mappings, verify ALL of the following. Report pass/fail per item.
Format and version compliance
Mapping values are version-pinned (ADR-027): every value carries its framework's version token, except STRIDE, which is intentionally unversioned. Verify each value against the canonical pinned pattern for its framework in the authoritative style guide (risk-map/docs/contributing/framework-mappings-style-guide.md, "Identifier Enforcement" table) — read the patterns from the guide every time; do not restate them from memory or inline here. The guide is the single source for pattern and version-token conventions across every supported framework, including any framework registered after this skill was written — new frameworks are a recurring, actively-requested change (see open issues against frameworks.yaml), so nothing about the pinned-pattern set may live in two places.
Flag any value whose form or version token does not match the guide's current pinned pattern.
(This skill AUDITS existing mappings across the corpus; to SELECT mappings while authoring a single control or risk, use the mapping-selection skill.)
Structural compliance
No parent + sub-technique collisions: If AML.T0010.002 is used, AML.T0010 must NOT appear on the same entity
No technique/mitigation crossover: Risks use only AML.T####; controls use only AML.M####
applicableTo respected: Each framework is only used on entity types listed in its applicableTo (per style guide table)
Selectivity compliance
Soft limit (4 per framework): Flag any entity with more than 4 mappings in a single framework. Not a hard error, but requires justification.
Direct relevance: Each mapping should be defensible with a one-sentence rationale. Flag mappings where the connection is "related" rather than "directly relevant."
Term and identifier verification
Two distinct checks — do not conflate them. The first is structural and runs at whatever scope the audit is already covering; the second is a live, per-entity lookup that is never an automatic side effect of the first.
Verify identifiers are well-formed against the style guide's patterns. For MITRE ATLAS, technique/mitigation IDs match the pinned regex. For OWASP, the category title matches the guide. For ISO 22989, role names match the controlled vocabulary. This is a structural check — no live lookup — and applies at whatever scope the audit already covers, whole-corpus included, the same as the structural-compliance and selectivity checks above.
Live-verify identifier currency (search the web) when a specific identifier's currency is actually in question: proposing a new mapping value, or auditing a single targeted entity (per Scope) whose identifiers you are specifically confirming. This is a heavier, per-entity check. It is not run automatically across every entry as a side effect of a whole-corpus audit — a whole-corpus audit's job is the structural/selectivity/overlap sweep above, not a live web-search of every existing mapping in the file. If comprehensive liveness verification of the full corpus is genuinely wanted, invoke it as its own explicit, separately-requested pass — never as an implicit consequence of the routine audit.
Coverage analysis (for whole-corpus scope only)
Count entities with vs. without mappings per entity type
Identify unmapped entities where obvious framework matches exist
Note underutilized frameworks (e.g., NIST AI RMF, OWASP LLM08)
Note STRIDE categories never used
Output format
For targeted audits (single entity), provide:
Current mappings listed per framework
Pass/fail on each checklist item
Any recommended additions or removals with rationale
