Update: closed the audit→fix loop today.

CORE now prints the exact context build command for every audit violation:

  core-admin context build \
      --file src/will/agents/coder_agent.py \
      --task code_modification \
      --output var/context_for_claude.md

Run audit → get 65 actionable commands → paste output to Claude → fix. Zero manual translation between "what's broken" and "what to show the AI."

The system also audited the files we wrote today and flagged its own modularity debt. Law outranks intelligence, including ours.

Six-dimension self-assessment: Dim 1: Constitutional health (full 81-rule audit) Dim 2: Semantic landscape (Qdrant core_capabilities) Dim 3: Knowledge gaps (DB — intent/vector/signature coverage) Dim 4: Structural health (symbol inventory + layer distribution) Dim 5: Change context (git history + hotspot files) Dim 6: Intent drift (DB + on-demand embeddings → cosine distance)

LLM reasoning produces prioritised root-cause clusters. Campaign persisted as parent/child Tasks in PostgreSQL. Escalations flagged for human .intent/ amendment.

Constitutional bounds respected throughout — never writes to .intent/

- Constitutional safety (proven at A2) - NEW: Universal pattern that closes all loops (INTERPRET → ANALYZE → STRATEGIZE → GENERATE → EVALUATE → DECIDE) - Demo: CORE can explain itself through its own architecture - 70-80% autonomous code generation success - 100% local, MIT licensed - Looking for feedback on component architecture

What keeps an AI from drifting into unmaintainable chaos?

I started building the answer four years ago: CORE — open-source framework for constitutionally aligned AI-driven development.

Key pieces: - Immutable Constitution: 10-min readable, copy-paste into any project - ConstitutionalAuditor: blocks drift instantly - Human Quorum: required for edge cases (GitHub/Slack ready) - Self-Healing Loops: `core-admin fix all` stays inside the rules - Mind–Body–Will architecture: fully modular & traceable

Alpha, MIT, 5-minute QuickStart.

Built for the post-Opus world: AI codes fast; CORE ensures it codes right.

Docs: https://dariusznewecki.github.io/CORE/ Worked example: https://github.com/DariuszNewecki/CORE/blob/main/docs/09_WOR...