Docs exposed document blocks to the agent as text only, so an image block
read as 'an image' with no way to inspect its pixels. The analyze_media
tool already existed for Slides; port the capability to Docs:
- media-tools: accept data: URLs in loadMediaReference (pasted document
images live as base64 data URLs in image blocks)
- docs main/preload/shared ipc: expose ai:analyze-media to the renderer
- docs skill/tools: add the analyze_media tool (blockIndex or url), gated
on mediaAnalysisAvailable like generate_image
- protocol: annotate empty image blocks in the block list with a pointer
to analyze_media so the model discovers the tool
- i18n(ai): aiSumAnalyzeMedia/aiSumAnalyzeMediaDone across all locales