Uploaded image descriptions (via image_vision_analysis) are accurate but too long — multi-paragraph, listing every visible detail (colors, background objects, textures). Held/committed assertions should carry a brief summary description, not an exhaustive one.
Tune the vision skill's prompt to produce a short, one-paragraph summary rather than an exhaustive description. Same check for CR-2026-133's text-summarization output — confirm its output is also brief, not just gated by length threshold.
No conflict. Description length is not a seed commitment — specification-level tuning.
No change to which files get descriptions, no change to the confirm/commit flow.