> Preserved source method from `amplifier-bundle-ergonomics` at `8a6b9a0a5189b5572ea5d5c933f7956227a711fb`. Original examples and runtime claims are historical; the canonical entry point defines current scope.

---
name: human-agent-authority-map
description: Allocate decision and action authority between humans and agents according to consequence, reversibility, uncertainty, expertise, preference, and recovery capability.
---

# Human-agent authority map

For each meaningful agent action record:

- user intent and authorized scope
- information visible to the agent and user
- decision owner
- execution owner
- uncertainty and consequence
- reversibility and recovery time
- preview, approval, intervention, undo, retry, escalation, and completion evidence
- retained memory and deletion control

Choose an authority pattern: inform, suggest, co-plan, require approval, delegated action with monitoring, or prohibit automation. Scale friction with consequence and reversibility. Optimize for appropriate reliance, not maximum autonomy or confirmation frequency.