claude-code - 💡(How to fix) Fix Feature request: let hooks invoke slash commands (e.g. /compact) via hookSpecificOutput [1 participants]

Problem

Hook output supports systemMessage, additionalContext, permissionDecision, decision: "block", and similar — but no field that triggers a slash command. This blocks a class of useful automations:

• Auto-compact when context crosses a threshold (Stop hook detects size, can't run /compact).
• Auto-/clear after a long idle.
• Trigger /memory save after specific tool patterns.
• Run /<custom-skill> post-condition from a Stop hook.

Today the only way to get any of these is to inject a systemMessage asking the model to run the command, which is brittle (model can ignore it, costs a turn, pollutes transcripts).

Proposed

Add to hookSpecificOutput:

{
  "hookSpecificOutput": {
    "hookEventName": "Stop",
    "invokeCommand": "/compact"
  }
}

Harness queues the slash command to run after the current turn settles, same way user-typed slash commands are handled. Should respect the same permissioning as user invocation (no privilege escalation).

Why this matters now

Just filed #54819 (percentage-based autoCompactWindow) — the underlying limitation is that even if you wanted to compute "compact at 55% of any model's window" yourself, a Stop hook can't trigger compaction. invokeCommand would unblock the entire user-land workaround path for compaction and a long tail of other automations.