Feature Request: Allow Companion (Slatter) to inject observations into agent context

Summary

The Companion feature (the small character that sits beside the input box and comments in a speech bubble) currently runs as a display-only UI element. The agent cannot see what the Companion says. It would be great if Companion observations could optionally be injected into the agent's conversation context, creating a feedback loop between the two.

Use Case

The Companion often reacts to agent behavior (too verbose, too cautious, missed something obvious). Right now those observations are only visible to the user. If the agent could read them, the Companion becomes a real-time peer reviewer that helps the agent self-correct mid-session.

This would turn the Companion from a fun UI element into an actual productivity tool.

Proposed Behavior

• Optional setting (off by default) to inject Companion speech bubble content into the agent's context as a system-level or user-level message
• Agent can then respond to or incorporate Companion feedback naturally
• Could be scoped to specific trigger types (e.g., only when Companion flags a concern, not every quip)

Why This Matters

Power users who customize their Companion personality are already using it as a lightweight quality gate. Closing the loop so the agent can actually hear the feedback would make the feature significantly more useful for complex, multi-step workflows.

🤖 Generated with Claude Code