Preserve original resolution images alongside resized versions

[srinathh]

Problem

The image-vision skill currently resizes all incoming images to max 1024×1024 JPEG and discards the original. The agent only has access to the resized version.

This is limiting for downstream workflows that need full-resolution images — for example:

• Passing to Google Gemini or other vision models that handle higher resolution
• Saving originals to cloud storage (GDrive, S3)
• OCR on detailed documents or receipts

Proposed Solution

Save both the original and resized versions:

attachments/
  img-1711234567890-abc1.jpg              # resized (max 1024px, JPEG 85%) — sent to Claude
  img-1711234567890-abc1-original.png     # original resolution, original format

• The resized image continues to be sent as the multimodal content block to Claude (keeps token usage reasonable)
• The original is preserved on disk and the agent is informed of its path via a text message
• The agent can then use the original for any downstream action that benefits from full resolution

Changes

• src/image.ts — detect original format via sharp metadata, save both versions, update interfaces and parser
• container/agent-runner/src/index.ts — inform agent of original file paths after multimodal blocks
• src/container-runner.ts, src/index.ts — pass originalRelativePath through the pipeline

Notes

• Twilio (and possibly other channels) re-encode images on delivery, so sending the original back via WhatsApp still results in some quality loss. The real value is for agent-side processing (API calls to other models, file storage, etc.)
• Original format is detected and preserved (PNG stays PNG, WebP stays WebP, etc.)

🤖 Generated with Claude Code