Skip to content

CLI Contract — Assistive AI (v1) (Authoritative)

This contract prevents AI from becoming authoritative.

Hard rules

  • AI is advisory only; must emit authority: advisory
  • AI commands never modify verdicts, exit codes, or artifacts
  • AI commands never run labs or call runtime
  • AI commands exit 0 unless usage/artifact error (exit 2)

Commands

cassian ai

Inputs: - explicitly invoked operator input only - supporting topology and artifact context when available

Behavior: - advisory only - non-authoritative - never modifies verdicts, exit codes, or artifacts - never runs labs or calls runtime

Outputs: - explanations, summaries, and bounded guidance consistent with shipped behavior - must not invent facts - must remain outside the authority chain

Provider boundary

  • AI is opt-in:
  • enabled only when explicitly invoked and provider is configured
  • BYO key via environment variables only
  • If provider missing: print deterministic scaffold output; exit 0