[BUG] PreToolUse/PostToolUse warn hook systemMessage silently dropped without hookSpecificOutput

Summary

systemMessage returned from a PreToolUse or PostToolUse hook without hookSpecificOutput is silently dropped — not shown to the user in the terminal and not injected into the model context. This makes warn action rules in the hookify plugin completely invisible.

Environment

• Claude Code (latest)
• Platform: Windows 11
• Hookify plugin via claude-plugins-official marketplace

Steps to Reproduce

1. Create a PreToolUse hook that returns only systemMessage (no hookSpecificOutput):

``json
{ "systemMessage": "⚠️ Warning: dangerous command detected!" }
``

1. Trigger the matching tool call (e.g., rm -rf /tmp/test with a rm\s+-rf pattern rule).
2. Observe: the command executes, no warning appears in the terminal, and no context is injected into the model.

Expected Behavior

• systemMessage should render visibly in the terminal UI (e.g., ⎿ PreToolUse:Bash says: ⚠️ Warning...)
• Or at minimum, the message should be injected into the model's context so Claude can act on it

Actual Behavior

• Message is silently dropped
• User sees nothing; model sees nothing
• The command proceeds without any indication the hook fired

Root Cause

Claude Code appears to require hookSpecificOutput to be present for tool hook events in order to process (and display) a response. Without it, the JSON output from the hook is parsed but the systemMessage has no rendering path.

For block actions, hookSpecificOutput.permissionDecision = "deny" triggers the denial UI. For warn actions, there is no equivalent trigger, so the message is orphaned.

Fix (Verified Working)

Adding hookSpecificOutput with permissionDecision: "allow" and additionalContext resolves both issues — the model receives context injection and (pending confirmation) the terminal renders systemMessage:

return {
    "hookSpecificOutput": {
        "hookEventName": hook_event,   # "PreToolUse" or "PostToolUse"
        "permissionDecision": "allow",
        "additionalContext": combined_message
    },
    "systemMessage": combined_message
}

After applying this fix, additionalContext is confirmed injected into model context as PreToolUse:Bash hook additional context: ...

Related Issues

• #20747 — Hookify warnings don't reach the model (only shown to user) — same root cause, opposite framing
• #32624 — Hook systemMessage display prefix verbosity — confirms systemMessage does render for SessionStart events
• #17356 — permissionDecisionReason not displayed in permission prompt for PreToolUse 'ask' decisions — related display gap

Suggested Fix for Hookify Plugin

Update plugins/hookify/core/rule_engine.py evaluate_rules() to include hookSpecificOutput for warn actions on tool events. PR forthcoming if helpful.

🤖 Generated with Claude Code