claude-code - 💡(How to fix) Fix Allow hooks to spawn Monitor or invoke tools directly [2 comments, 2 participants]

Feature Request: Allow hooks to spawn Monitor (or invoke tools)

Problem

Today, hooks can only output text that becomes system reminders. If a hook detects a meaningful event (e.g. git push to a branch with an open PR), the only way to start background monitoring is through an indirect pipeline:

1. Hook outputs a reminder
2. Claude reads the reminder
3. Claude invokes a skill
4. The skill tells Claude to call Monitor

This works but has friction: it depends on Claude interpreting the reminder correctly every time, it consumes context window, and the user has no guarantee the Monitor actually starts.

Proposed Solution

Allow hooks to return a structured response that directly spawns a Monitor (or more generally, invokes a tool). For example, a PostToolUse hook could return:

{
  "tools": [
    {
      "name": "Monitor",
      "input": {
        "command": "bash ~/.claude/skills/watch-pr-checks/scripts/poll-checks.sh 123",
        "description": "PR #123 CI checks",
        "persistent": true,
        "timeout_ms": 3600000
      }
    }
  ]
}

Use Case

I built a watch-pr-checks skill that monitors GitHub CI checks for the current PR via Monitor. A PostToolUse hook detects git push and gh pr create and should automatically start the monitor. Today the hook can only suggest that Claude start the monitor; it can't guarantee it.

This pattern generalizes: any hook that wants to start a background watcher, trigger a build, or kick off a long-running process after detecting a specific event would benefit from direct tool invocation.

Alternatives Considered

• Hook reminder → Skill → Monitor: Current approach. Works but is indirect and fragile.
• Hook starts a background process directly: The process runs outside Claude's context, so Monitor notifications don't reach the conversation.