tools-only Agent Calibration Reference
Prime agents with a boot message on creation, and recalibrate after updates.
First Message
The first_message field sends a message to the agent on first creation only (never on updates). Used to make agents internalize their role.
agents:
- name: support-agent
first_message: "Review your persona block and confirm you understand your role."
The message is sent asynchronously after creation.
Recalibration
After updating agent config, recalibrate to re-prime agents:
# Recalibrate all changed agents
lettactl apply -f fleet.yaml --recalibrate
# Custom message
lettactl apply -f fleet.yaml --recalibrate --recalibrate-message "Review updated guidelines."
# Scope by tags
lettactl apply -f fleet.yaml --recalibrate --recalibrate-tags "role:support"
# Scope by pattern
lettactl apply -f fleet.yaml --recalibrate --recalibrate-match "support-*"
# Fire-and-forget
lettactl apply -f fleet.yaml --recalibrate --no-wait
SDK
await ctl.bulkSendMessage('Review your updated configuration.', {
pattern: 'support-*',
});
Options
| Flag | Description |
|---|---|
--skip-first-message |
Skip first_message on creation |
--recalibrate |
Re-send calibration messages |
--recalibrate-message <msg> |
Custom calibration message |
--recalibrate-tags <tags> |
Scope by tags (AND logic) |
--recalibrate-match <pattern> |
Scope by glob pattern |
--no-wait |
Fire-and-forget |
Retraining vs Recalibration
- Recalibrate: Send a message in-place, preserves conversation history
- Retrain: Delete and redeploy for a fresh start (
lettactl delete agent <name>thenlettactl apply)