SuperAPI AI Directory

hooks

General & other

View repository

Updated 10-15 23:49

About this skill

Hooks - Lifecycle hook system

The Task plugin provides 2 official hooks for session-level event handling. Internal lifecycle events (task start/completion/iteration) are handled through the Agent's internal logic and do not use the hooks system.

Available Hooks

SessionStart

Trigger: on session start/resume. Purpose: initialize environment variables/load configuration/set up resources.

{"SessionStart": [{"hooks": [{"type": "command", "command": "echo 'export ...' >> \"$CLAUDE_ENV_FILE\"", "async": true}]}]}

SessionEnd

Trigger: on session end. Environment variables:SESSION_ID. Purpose: clean up temporary files/archive logs/generate reports.

Log:.claude/logs/task-hooks-{session_id}.jsonl

Migration notes (v0.0.184)

Removed 6 hooks not officially supported (TaskStart/IterationStart/IterationEnd/TaskComplete/TaskFailed/CheckpointSave). Alternatives:

Original Hook New Solution
TaskStart Loop skill output[MindFlow] 任务启动
IterationStart/End Loop skill iteration log
TaskComplete Finalizer generates report
TaskFailed Adjuster records failure
IterationEnd (metrics) Observability skill
CheckpointSave Checkpoint skill log

Hook Development Guide

  • Input: environment variables (SESSION_ID, etc.) + stdin JSON
  • Output: stdout log | stderr error | exit 0=success, 2=block, other=non-blocking
  • Best practices: execute in <1 second | async:true | robust error handling | idempotent | default timeout 600s

Troubleshooting

Not triggered: check that the hook name is in the official list/plugin.json configuration/script is executable/path is correct. Failure: check syntax/environment variables/stderr/timeout/permissions.