feat(feishu): add inbound image message support

[shikihane]

Description

Add inbound image message support for the Feishu channel. When users send images (or images with text) via Feishu, the channel now downloads and stores them as media:// refs for the vision pipeline to process.

Depends on: #1020 (vision pipeline — resolves media:// refs to base64 data URLs for the LLM)

Changes

pkg/channels/feishu/feishu_64.go:

• Pure image messages (MsgTypeImage): extract image_key from {"image_key":"..."} content, download via MessageResource API
• Rich text post messages (MsgTypePost): parse {"content":[[{"tag":"img","image_key":"..."}]]} format to extract all embedded images
• downloadImage(): uses MessageResource.Get(messageID, fileKey, type="image") — the correct API for downloading user-sent images (Image.Get only works for bot-uploaded images)
• No hardcoded extensions: filenames use image_key instead of .jpg — MIME detection is handled downstream by h2non/filetype magic bytes
• MediaStore integration: downloaded images stored via store.Store() → media:// ref → passed to agent via HandleMessage()

pkg/channels/feishu/feishu_64_test.go (new file):

• extractFeishuImageKey tests (valid key, empty, invalid JSON, missing field)
• extractFeishuMessageContent tests (text, image, nil message)

Data flow

Feishu image/post message
  → handleMessageReceive() detects MsgTypeImage or MsgTypePost
  → extractFeishuImageKey() / extractFeishuPostImageKeys()
  → downloadImage() via MessageResource.Get API
  → MediaStore.Store() → media://uuid ref
  → HandleMessage(..., mediaPaths, ...) → InboundMessage.Media
  → (vision pipeline resolves to base64 for LLM)

E2E verification

Tested on Radxa Cubie A7A (arm64) with Feishu channel + Dashscope LLM (kimi-k2.5). Sent image via Feishu, LLM correctly identified image content.

Type of Change

• ✨ New feature (non-breaking change which adds functionality)

AI Code Generation

• 🛠️ Mostly AI-generated (AI draft, Human verified/modified)

Test Environment

• Hardware: Radxa Cubie A7A (arm64, 4GB RAM)
• OS: Debian 11 (bullseye)
• Model/Provider: Dashscope (kimi-k2.5, vision-capable)
• Channel: Feishu (WebSocket mode)

Checklist

• My code follows the style of this project
• I have performed a self-review of my own changes
• Backward compatible — non-image messages unaffected
• All existing tests pass (make test)

[shikihane]

✅ Implementation Complete & Tested

The full vision pipeline is now working end-to-end on our deployment:

What's been implemented (on dev branch)

1. Feishu inbound image download (commits 9527e95, 8912dbb)

   • Detects MsgTypeImage, extracts image_key
   • Downloads via MessageResource.Get() API (not Image.Get() — that only works for bot-uploaded images)
   • Stores as media://uuid refs in MediaStore

2. Agent pipeline Media support (commits 1a55a8a, 09e0220)

   • Message.Media field propagation through context builder
   • serializeMessages() converts Media to OpenAI vision format

3. Media resolution to base64 (latest commit, not yet pushed)

   • resolveMediaRefs() converts media:// refs to data:image/...;base64,... URLs before LLM call
   • Session saves lightweight media:// refs (not bloated base64)
   • Feishu sets ContentType in MediaMeta for proper MIME inference

Test results

Hardware: Radxa Cubie A7A (arm64, 4GB RAM)
Model: qwen3.5-plus via coding.dashscope.aliyuncs.com/v1 (supports vision)
Channel: Feishu WebSocket

Screenshot attached below — user sent a dog photo, bot correctly identified it as "好可爱的柴犬！🐕" with detailed description.

Next steps

Ready to rebase clean commits onto a feat/vision-feishu branch from main for PR submission. Should this PR include:

• Just Feishu inbound image support, OR
• Full vision pipeline (Feishu + agent resolution)?

[shikihane]

Closing this PR — the inbound image support is now fully covered by #1020 (Vision Pipeline V2) which has been merged. Thank you!