• Intent doesn't only live in a spec.
    • Intent may remain in JIRA tickets, PRDs, and Design docs, but ultimately the most important decisions are Your prompts
    • SPEC + Intent => Acceptance Criteria
  • The AI Slop Register
    • Shared organizational rules that AI misses (Invariants)
  • Acceptance Criteria (SPEC + Intent) + Invariants => Test Plan
  • The structure we have now seems pretty well designed? SPEC and Intent are each organized within the SPEC document, while Invariants are being handled through custom lint rules. From here, wouldn't it be enough to turn Intent and Custom Lint rules into a compact, well-structured form of knowledge that LLMs can access more easily?