AI Asset Governance

This document defines how project-critical AI assets are versioned, mirrored, and recovered.

Goal

No essential AI-assisted delivery behavior should depend solely on a local, unversioned Cursor configuration.

Canonical sources

• docs/ai-governance/*.md
  • human-readable canonical specs for agent roles, authority, workflow, and recovery rules
• docs/prompts/**
  • canonical product/runtime prompt behavior
• docs/rebuild/**
  • recovery playbooks for reconstructing tooling from docs

Mirror assets

The following live assets are operational mirrors, not the only source of truth:

• ciq-mvp2-app/.cursor/agents/*.md
• ciq-mvp2-app/.cursor/rules/*.mdc
• ciq-mvp2-app/.cursor/skills/*/SKILL.md

Required governance rules

1. Every project-critical agent, rule, or skill must have a matching canonical doc or section in this repo.
2. If a .cursor asset changes, update the canonical docs in the same change.
3. If the live asset cannot be committed, the docs must still contain enough information to recreate it faithfully.
4. Machine-specific configuration outside the repo must document:
   • why it is external
   • what fields are project-critical
   • how to recreate and validate it

Machine-specific assets

Examples:

• ~/.cursor/mcp.json
• global Cursor-managed skills under ~/.cursor/skills-cursor/

These may remain machine-local, but their project-relevant configuration must be documented in docs/ai-governance/** and docs/ops/**.

Drift policy

If a conflict exists between:

• docs in ciq-mvp2-docs
• live .cursor assets
• remembered local setup

the docs repo must be treated as the human-reviewable source of truth, then the live assets must be corrected or regenerated to match it.