[Opportunity] Kimi CLI — Hooks & Lifecycle Events System

[howardpen9]

Target Project

Kimi CLI (MoonshotAI/kimi-cli)

Related Issue(s)

• Kimi #1313 — Feature Request: Add Hooks System

What

Implement a hooks system allowing users to run custom scripts at key lifecycle points:

• PreToolUse / PostToolUse (before/after tool execution)
• PreMessage / PostMessage (before/after sending to LLM)
• Notification (task completion, errors)
• Session lifecycle (start, resume, end)

Why

• Explicit user request in Kimi #1313
• Enables: desktop notifications, logging, safety checks, custom integrations
• Claude Code has hooks but with many bugs (#31469, #31440) — we can learn from their mistakes

Prior Art

Claude Code Hooks (Lessons Learned)

What works:

• JSON config in settings files
• Pattern matching on tool names
• Shell command execution with JSON context on stdin

What breaks (avoid these):

• Hooks not propagated to subagents (#31469)
• Non-deterministic execution order (#31440)
• No timeout on hook scripts (can hang the agent)

Implementation Plan

1. Define hook points enum (PRE_TOOL, POST_TOOL, NOTIFICATION, etc.)
2. Add hooks config in kimi config file (YAML/JSON)
3. Hook runner: spawn subprocess with context on stdin, capture stdout
4. Hook result handling: allow hooks to block/modify/log
5. Add timeout per hook (default 10s)
6. Propagate hooks to subagents
7. Built-in hooks: desktop notification on task complete

Estimated Effort

Medium (1-2 weeks)

Status

• Research complete
• Implementation started
• PR submitted
• PR merged