dtpr-comprehension-audit

The standalone comprehension-tier skill. Grades DTPR content against the comprehension rubric and returns a structured findings block plus a one-paragraph summary. Grading only — it does not modify schema content, draft elements, or change category structure.

The four authoring skills that emit schema:new proposals (dtpr-datachain-structure, dtpr-category-audit, dtpr-element-design, and dtpr-translate) inline a first-pass Comprehension check in their output. Use this skill for a deeper second pass, a re-grade after the rubric evolves, or a one-off grading without schema work.

When to fire

• The user asks to grade, audit, or check a DTPR element, category, datachain-instance, or pasted content for public comprehension.
• The user uses "comprehension audit", "is this clear", "how readable is X", or "does this land for a non-expert".
• A schema-tier skill's inline Comprehension check needs a deeper standalone second pass.
• The user pastes YAML, JSON, or markdown and asks whether it would be understood by a non-technical person encountering an AI system — in an app, on a website, in a vehicle, on a kiosk or sign, or any other surface where DTPR content might appear.
• The user wants to re-grade after the rubric's rubric_version has changed.

Trigger phrases

"grade this element", "is this category clear to the public", "comprehension audit", "check this datachain for public understanding", "how readable is X", "does this land for a non-expert".

Workflow

Phase 0 — Identify the target. Classifies the input as an element id, category id, pasted element YAML, datachain-instance JSON, arbitrary markdown, or a mix. Asks a structured clarification question if ambiguous.

Phase 1 — Load the target content. Uses get_element, list_elements with query, list_categories, get_schema, or get_elements as the input type requires. Pasted content is graded directly without an MCP read.

Phase 2 — Read the rubric and template. Reads references/comprehension-rubric.md and references/comprehension-block-template.md at runtime. Captures the rubric_version from the rubric's frontmatter.

Phase 3 — Corpus lookup. Reads the research corpus with tags like category:<id>, element:<id>, or concept:<slug>. Writes a corpus entry only when the finding is a non-obvious insight worth compounding.

Phase 4 — Apply the rubric item by item. One verdict per item (pass / fail / partial / n/a) with a one-line reason that names the specific phrase, element, or pattern that drove the verdict.

Phase 5 — Emit findings. The inline block (exact template shape) plus a one-paragraph summary that names the top fix(es) and routes the user to a sibling skill when a gap needs schema-level fixing.

Output

1. The Comprehension check block (verdicts + reasons + Rubric version: trailer).
2. A one-paragraph summary that names the top one or two fixes and any sibling-skill handoff.

Non-goals

Does not modify schema content — edits happen in the schema-tier skills and are committed by humans after running schema:new. Does not replace the inline Comprehension block that the schema-tier skills emit as part of their proposals. Does not translate or localize content — locale coverage is graded as a rubric item but actual translation is out of scope.

Source

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