Background

While trying to implement automatic Slack thread context injection for our AI agents, we consulted the hooks documentation (docs/hooks.md). The doc lists message:received under "Future Events", which led us to believe this functionality was not available.

However, after reverse-engineering the implementation (dist/loader-n6BPnYom.js), we discovered that many hooks are already implemented but not documented:

Implemented but Undocumented Hooks

Key Discovery: before_agent_start Hook

This hook is particularly powerful:

async function runBeforeAgentStart(event, ctx) {
  return runModifyingHook("before_agent_start", event, ctx, (acc, next) => ({
    systemPrompt: next.systemPrompt ?? acc?.systemPrompt,
    prependContext: ...
  }));
}

Capabilities:

• event.prompt — User's message
• event.messages — Session conversation history
• ctx.sessionKey — Session key (can identify channel/thread)
• ctx.agentId — Agent ID
• ctx.workspaceDir — Workspace directory
• ctx.messageProvider — Message provider (e.g., "slack")
• Returns prependContext to inject context before the prompt
• Returns systemPrompt to modify system prompt

This allows fetching external context (e.g., Slack thread history) and injecting it before the agent processes the message.

Request

Please update docs/hooks.md to document these implemented hooks. Other users may be giving up on features that are actually available.

Environment

• OpenClaw version: 2026.2.6-3
• Installation: npm global (homebrew)