Summary

• When users attach images (PNG, JPEG, WebP, GIF) to their Slack messages mentioning the bot, the images are now automatically extracted, downloaded via the chat SDK's authenticated fetchData(), uploaded to R2 in the Cloud Agent attachments bucket, and forwarded to the Cloud Agent session via PrepareSessionInput.images.
• The bot agent LLM is informed when images are present so it can reference them in the prompt it constructs for the Cloud Agent (e.g. "implement the design shown in the attached screenshot").
• Image processing is non-fatal — if extraction or upload fails for any image, the bot logs the error to Sentry and continues without it. Respects existing limits (max 5 images, max 5MB each).

Visual Changes

Reviewer Notes

• The new extractAndUploadImages utility (apps/web/src/lib/bot/images.ts) uses the chat SDK's Attachment.fetchData() which handles Slack's url_private authentication automatically via the bot token — no manual token handling needed.
• Images are uploaded directly to R2 server-side using PutObjectCommand (unlike the browser flow which uses presigned URLs). This is analogous to how cli-sessions.ts does direct R2 uploads.
• The R2 key structure follows the existing convention: {userId}/cloud-agent/{messageUuid}/{imageId}.{ext}.
• No changes to the Cloud Agent worker are needed — it already supports the images field in prepareSession.

Built for Mark IJbema by Kilo for Slack