Eval Criteria: ai-layer-kit/CLAUDE-root.md

Eval Criteria: ai-layer-kit/CLAUDE-root.md

Created: 2026-05-26 Target: production-lines/digital-talent/templates/ai-layer-kit/CLAUDE-root.md Purpose: Root CLAUDE.md template loaded by every delivered talent at session start — establishes identity, execution rules, kit awareness, domain routing, and safety floor in the leanest form possible.

Questions

Token economy

  1. Is the file under 60 content lines (excluding code fences)?
  2. Is every section ≤ 12 lines?

Identity & bootstrapping

  1. Within the first 10 lines, does the file establish role + client + language?
  2. Are the 4 execution rules present with their canonical labels (Think before producing / Simplicity first / Surgical changes / Goal-driven execution)?
  3. Is the "do not edit" warning placed before (not after) the execution rules?

Routing & integration

  1. Does the file explain when to load a domain-specific CLAUDE-{domain}.md (vs. staying on root)?
  2. Are all file paths referenced in this file actual files that the kit creates or live files the assembly produces? (no invented paths)
  3. Does the kit components table list exactly the 5 components from the AI-Layer Kit v1.0?

Placeholders

  1. Do all placeholders use the same {curly-brace} convention (no mixed syntax)?
  2. Does every placeholder have either an inline hint (e.g., "e.g., legal review") or an obvious sample value to guide Pablo when filling?

Safety & actionability

  1. Does the safety floor contain at least 3 non-negotiable rules?
  2. Are the execution rule and safety floor statements written as imperatives the LLM can act on (verb-first or clear directives)?