Enhancement: separate personality bubble text from evidence-based technical observations in buddy_observe

[terpjwu1]

Problem

buddy_observe currently renders the visible bubble directly from templateFallback. That means the same deterministic fallback path is responsible for both:

• personality flavor
• what reads like technical observation

In practice, this blurs the line between:

• generic companion flavor
• an evidence-backed code comment

Current behavior

When no max-mode finding is present, the user still sees text that can sound like a concrete observation, even though it is only coming from fallback template pools.

Request

Split the output model conceptually and in implementation:

1. A generic in-character reaction bubble that is always safe to show
2. A separate technical observation field that is only surfaced when Buddy has actual evidence, such as:

• a max-mode finding
• explicitly passed structured context from the host
• some future diff-aware input path

Why this matters

This would preserve the product's personality while reducing false-positive-sounding code comments. It would also make the contract clearer to users: Buddy is always allowed to emote, but it should only make specific technical claims when it has grounds to do so.

Suggested implementation direction

• keep templateFallback for flavor
• gate specific technical observation text behind finding or another evidence-bearing field
• optionally surface both distinctly in the JSON response so hosts can render them differently