DTPR for AI
Taxonomy
Skills

dtpr-category-audit

Audit one category's element collection for coherence, overlap, and gaps.

The category-tier schema skill. Scope is exactly one category — does the element collection cover the expected territory without overlap? Coherence = each element occupies a distinct slot. Overlap = two elements converge on the same meaning. Gaps = a risk or scenario no element captures.

When to fire

  • The user names a specific category and asks whether it is complete for a class of systems.
  • The user suspects two elements within a category are redundant.
  • The user wants a walkthrough of a category for multi-party responsibility, generative AI, modern appeals, or similar.
  • The user references trigger phrases that scope to one category (e.g., ai__risks_mitigation, ai__accountable).

Trigger phrases

"does DTPR cover", "what's missing from the schema", "accountable deep-dive", "is the ai__risks_mitigation category complete", "audit this category", "elements overlap in DTPR's X category", "gaps in ai__decision".

Workflow

Phase 0 — Accept and confirm the category. Fuzzy-matches typos against list_categories and confirms before proceeding.

Phase 1 — Fetch the category and its elements. list_categories for metadata, list_elements scoped to the category, get_elements for bulk bodies.

Phase 2 — Overlap detection. For each element, calls list_elements with query=<element.title> + same category_id to surface relevance-ranked neighbors. Flags converging pairs.

Phase 3 — Corpus lookup. Tags typically include category:<id>, concept:<domain>, framework:<name>, standard:<name>. Dispatches a researcher on miss when Task is available.

Phase 4 — Gap analysis. Enumerates risks and scenarios from the corpus and first-principles; checks each against the current elements; flags unmatched concepts as gaps.

Phase 5 — Emit the audit using the Scenario / Coverage map / Overlap pairs / Gaps / Proposed changes {Add/Edit/Retire} template.

Phase 6 — Inline the Comprehension check block.

Phase 7 — Emit the schema:new handoff and recommend a sibling when relevant.

Output

A category-scoped audit (coverage map, overlap pairs, gap list, proposed element additions/edits/retirements), the inline Comprehension check, the schema:new command line, and any sibling recommendation.

Non-goals

Does not change the datachain-type shape — route to dtpr-datachain-structure. Does not draft full element YAML — route to dtpr-element-design. Does not describe a specific AI system — route to dtpr-describe-system.

Source

plugin/dtpr/skills/dtpr-category-audit/SKILL.md in the repository.

dtpr-datachain-structure

Critique or propose changes to the datachain-type shape itself — categories, requirements, retirement.

dtpr-element-design

Draft, edit, or retire one element — its title, description, variables, and symbol disposition (reuse / hand-off to dtpr-symbol-design / iterate).