OpenClaw Hooks

OpenClaw has two distinct hook systems: internal gateway hooks and plugin hooks.

Internal (Gateway) hooks

Event-driven shell scripts triggered by agent and command lifecycle events.

Hook	Trigger
`agent:bootstrap`	While building bootstrap files, before system prompt is finalized. Use to add/remove bootstrap context files.
`/new`, `/reset`, `/stop`	Command-level events.

Extension points inside the agent/tool lifecycle and gateway pipeline. Registered via the plugin API.

Hook	Phase	Purpose
`before_model_resolve`	Pre-session (no messages)	Deterministically override provider/model
`before_prompt_build`	After session load (with messages)	Inject `prependContext`, `systemPrompt`, `prependSystemContext`, `appendSystemContext`
`before_agent_start`	Legacy	Compatibility alias; prefer explicit hooks above
`before_agent_reply`	After inline actions, before LLM call	Claim turn → return synthetic reply or silence it
`agent_end`	After completion	Inspect final message list + run metadata
`before_compaction` / `after_compaction`	Compaction cycle	Observe or annotate
`before_tool_call` / `after_tool_call`	Tool execution	Intercept params/results
`before_install`	Skill/plugin install	Inspect scan findings, optionally block
`tool_result_persist`	Before transcript write	Synchronously transform tool results
`message_received`	Inbound message	Observe/modify incoming messages
`message_sending` / `message_sent`	Outbound message	Guard or observe outgoing messages
`session_start` / `session_end`	Session boundaries	Session lifecycle
`gateway_start` / `gateway_stop`	Gateway lifecycle	Startup/shutdown