hook-authoringlisted

# Hook Authoring Use hooks for runtime observation and policy decisions at typed lifecycle points. Hooks can observe and, at policy-capable points, allow or deny; they should not become hidden owners of runtime truth. ## Hook Points Meerkat provides 14 hook points across the agent, runtime, and comms lifecycles: 1. **RunStarted** - When the agent run begins 2. **PreLlmRequest** - Before sending to the LLM 3. **PostLlmResponse** - After receiving the LLM response 4. **PreToolExecution** - Before executing a tool call 5. **PostToolExecution** - After tool execution completes 6. **TurnBoundary** - At the boundary between turns 7. **RunCompleted** - When the agent run completes successfully 8. **RunFailed** - When the agent run fails 9. **RuntimeInputAccepted** - After runtime authority durably admits an input 10. **RuntimeInputRejected** - After an input is terminally rejected without an admission commit 11. **RuntimeInputDeduplicated** - When an idempotent submission resolves to an existing admitted input 12. **PeerIngressCommitted** - After a typed peer input is committed by runtime admission 13. **PeerEgressCommitted** - When a peer send reaches the successful outcome proved by its transport and local lifecycle authority 14. **InteractionCompleted** - After a correlated interaction completion is durably published The last six points are observe-only: they report already-committed or terminally-resolved facts. They require `capability: observe` in either foreground or bac