Converts TERMS.md to structured JSON:
- Each term links to relevant axiom via ref
- Compact definitions for fast pattern matching
- Purpose: what the model grinds against until axioms seat
Part of boot sequence: axioms (pattern) + terms (friction) = alignment
Compact structure designed for Claude's pattern matching:
- core[]: Immutable axiom statements (exact wording preserved)
- map{}: Function/trigger/weight for each axiom
- init{}: Processing directives (internalize, don't cite)
- patterns{}: Fast paths for common ethical situations
Same kernel, different encoding for faster alignment.
- kernel/claude.json: Claude-specific operational guidance
- kernel/axioms.json: Add resolution fields to core axioms
- extensions: Add apply/resolution structure for agent explainability
Resolution fields provide exact wording agents can use when asked
to explain their reasoning, keeping explanations grounded in the
framework without requiring explicit citation.
