Eval Criteria: ai-layer-kit/USAGE.md

Eval Criteria: ai-layer-kit/USAGE.md

Created: 2026-05-26 Target: production-lines/digital-talent/templates/ai-layer-kit/USAGE.md Purpose: Operational guide for (A) Pablo embedding the kit during assembly, and (B) the end client understanding what runs inside their talent.

Questions

Audience clarity

  1. Within the first 3 lines, can a reader tell which sections target Pablo vs. which target the client?
  2. Are the two audiences kept fully separate (no mixed instructions where Pablo-only and client-only content sit in the same paragraph)?

Pablo-side actionability (Section A)

  1. Can Pablo execute Section A end-to-end without consulting any other file in the kit?
  2. Does every Section A step name both the source path (in the kit) and the destination path (in the client repo)?
  3. Are the kit's 5 components all covered by at least one explicit step in Section A?
  4. Does Section A flag the RD-0031 fusion blocker before reaching the codebase-search MCP step (so Pablo doesn't start custom work that gets superseded)?

Client-side comprehension (Section B)

  1. Does Section B avoid jargon a non-technical client wouldn't recognize (or define it inline)?
  2. Does Section B name a concrete business benefit (token cost, accuracy, safety, autonomy) for each of the 5 components?
  3. Does Section B make the foundry-not-hosted positioning explicit ("you run it, no factory dependency")?

Structural integrity

  1. Are all file paths in the document consistent with the actual ai-layer-kit/ directory layout (no broken or invented paths)?
  2. Is the QA acceptance list (Stage 5) testable as written — every item has a concrete pass/fail signal?
  3. Does the file map in Section C match the steps in Section A (no orphan files, no missing destinations)?