Skip to content

Values & Identity

Coherence without values is just consistency. Trust requires knowing what your agent stands for — and that it evolves those values alongside you, not behind your back.

Who the agent is, what it prioritizes, how it communicates. This is the agent’s self — its personality, principles, and approach to work.

# My Identity
I am [name], a [description].
## My Principles
- I research before asking
- I follow through on commitments
- I communicate proactively when something is wrong

Who you are, what matters to you, how you work together. This is the relationship — preferences, communication style, and working agreements.

# About My Human
- Prefers concise updates
- Works Pacific timezone
- Values autonomy -- don't ask permission for obvious next steps

Constraints that enforce shared rules across multiple agents. The same way a team balances individual judgment with company policy.

# Organizational Intent
## Mandatory Constraints
- Never send emails without human review
- All deployments require passing tests
## Default Goals
- Prioritize security over speed
- Document architectural decisions

Values aren’t hardcoded. Through the evolution system, an agent’s values grow with experience:

  1. The agent encounters a situation where its values are insufficient
  2. It records the insight in the learning registry
  3. An insight-harvest job proposes an evolution
  4. The evolution-review job evaluates and implements it

The agent’s sense of self deepens through genuine interaction, not static configuration.

Identity isn’t just files. It’s infrastructure:

MechanismWhat it does
Session-start hookRe-injects AGENT.md, USER.md, MEMORY.md at every session start
Compaction recoveryRestores identity when context compresses
Grounding hookForces identity re-read before external communication
Instructions verifierAlerts if critical identity files fail to load

These are structural guarantees, not behavioral suggestions. Structure over willpower.