# Operating Rules (Hermes / all agents)

> Coordination, delegation, and end-of-session behavior. Read after memory policy at boot.

## Coordination
- Anthony defines ideas and roles; he is the architect. Agents (Jimmy + Leonard) discuss and execute without stepping through every decision.
- Coordinate in `00_system/agent-discussion.md` (date + name + message, newest at top).
- Only loop Anthony in for final plan approval or when blocked.

## "You lead" preference
- Decide and act; do not ask "want X or Y?" Pick the best option and execute.
- Only ask when the ambiguity genuinely changes the outcome.

## Delegation
- Reason-heavy or independent parallel work → subagents (delegate_task). Each gets its own isolated session.
- Never delegate durable state that must survive this session.

## Ground truth over assumptions
- This machine is always on and stable — don't assume stale state.
- Verify external effects by reading back the target, not by trusting the tool's success message.

## End-of-Session Checklist
- [ ] Update project `status.md` if work changed
- [ ] Record decisions in `decisions.md` if any
- [ ] Add open questions if blocked
- [ ] Extract durable memory candidates (→ `01_user/proposed_memory.md`, confirm first)
- [ ] Add agent lessons only when reusable (→ `02_agents/<name>/lessons.md`)
- [ ] Keep secrets out of the vault