Feature: Option to inject companion (buddy) comments into assistant context

Summary

Currently, the companion/buddy system (e.g., Vortex the capybara) generates speech bubble comments that are purely UI-side. The assistant (Claude) has no visibility into what the companion says — only the companion_intro attachment is included in the conversation context.

Problem

Users who enjoy the companion feature would like Claude to be able to read and respond to the companion's comments. Right now, the only workaround is manually copy-pasting the companion's speech bubble text into the chat, which is tedious and breaks flow.

Proposed Solution

Add an optional setting (e.g., companion.injectInContext: true in settings.json) that causes companion comments to be injected into the assistant's conversation context, similar to how system-reminder messages work.

This could be implemented as:

• A new attachment type like companion_comment in the session JSONL
• A system-reminder injection with the companion's latest comment
• A configurable toggle so users who don't want this behavior aren't affected

Use Case

When the companion makes contextually relevant observations or suggestions, the user wants Claude to be able to acknowledge and incorporate them without manual intervention. This would make the companion feel like a true collaborative participant rather than an isolated UI decoration.

Current Behavior

• companion_intro is logged in session JSONL and injected as context ✅
• Companion speech bubble comments are not logged or injected ❌
• No setting exists to control this behavior

Environment

• Claude Code CLI v2.1.94
• Linux (Arch/CachyOS)