AgentSkillsCN

Hook Authoring

为8个钩子点编写钩子，定义执行模式与决策语义。

SKILL.md
--- frontmatter
name: Hook Authoring
description: Writing hooks for the 8 hook points, execution modes, and decision semantics
requires_capabilities: [hooks]

Hook Authoring

Hook Points

Meerkat provides 8 hook points in the agent lifecycle:

  1. •RunStarted — When the agent run begins
  2. •PreLlmRequest — Before sending to the LLM
  3. •PostLlmResponse — After receiving 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

Execution Modes

  • •Foreground: Blocks execution, can modify/deny. Use for policy enforcement.
  • •Background: Runs concurrently, fire-and-forget. Use for logging, analytics.
  • •Observe: Receives events but cannot block. Use for monitoring.

Decision Semantics

Hooks return one of:

  • •Allow: Proceed normally
  • •Deny: Block the operation with a reason
  • •Rewrite: Modify the operation (e.g., rewrite tool arguments)

Patch Format

For PreToolExecution, hooks can rewrite tool arguments using JSON patches.

Failure Policy

  • •fail-closed: Hook failure blocks the operation (safer)
  • •fail-open: Hook failure allows the operation (more resilient)