Just tightened my ICM knowledge base / workspace and realized something that should have been obvious:
Most people write prompts.
Fewer people write identity.
So I made a reusable Operator Identity template for any workspace. It defines who the agent is, what loop it owns, how it should think, what it will not do, and which files are the source of truth.
The part that helped most was separating role from rules... Identity says what the Operator owns, while the rules say how work gets done.
- Operating Posture
- Quality Standard
- Core Work
- Boundaries
- Non-Negotiables
- Execution Sources
Curious how and others do this: Do you write a separate Identity for your agents, or do you keep everything inside one big instruction file? Anything else you include in Identity?