Skip to main content

AI Prompt Governance

Treat prompts as versioned runtime artifacts.

Versioning​

  • Store prompt changes in code with clear commit history.
  • Use explicit changelog notes for behavior-affecting updates.

Validation​

  • Evaluate new prompt variants on historical signal samples.
  • Compare quality distribution and approval rate shifts.
  • Check direction consistency and false reject/accept patterns.

Safety Controls​

  • Keep strategy direction constraints explicit in prompt logic.
  • Avoid vague instructions that allow unsupported behavior.
  • Gate rollout by environment (staging before production).

Operational Discipline​

  • Never change prompt and risk parameters blindly in one step.
  • Keep rollback-ready previous prompt variant.