fix(feishu): extract images from post (rich text) messages

[Vast-Stars]

Summary

• When users send text+image together in Feishu, the message type is post (rich text) instead of image. The image download logic only handled MsgTypeImage, so images embedded in post messages were silently ignored — the agent received raw JSON instead of actual image content.
• Add extractPostImageKeys() to parse img tags from post content
• Add extractPostText() to extract plain text from post content
• Handle MsgTypePost in downloadInboundMedia to download embedded images
• Handle MsgTypePost in appendMediaTags to add [image: photo] tag

Test plan

• Unit tests for extractPostImageKeys (text+image, multiple images, no images, invalid JSON)
• Unit tests for extractPostText (text+image, with title, only images, invalid JSON)
• Updated TestExtractContent for post message type
• Updated TestAppendMediaTags for post message type
• Manual test: send image-only message in Feishu → agent sees image (regression check)
• Manual test: send text+image message in Feishu → agent now sees both text and image

[nikolasdehor]

Well-structured fix for Feishu post (rich text) message image extraction. The changes are clean and well-tested.

Highlights:

1. extractPostImageKeys / extractPostText -- good separation of concerns. The JSON struct parsing is straightforward and handles the nested paragraph/element structure correctly.

2. Graceful degradation -- extractPostText falls back to rawContent on JSON parse error, while extractPostImageKeys returns nil. This asymmetry is intentional and correct: for text, showing raw content is better than nothing; for images, silently skipping malformed data avoids passing garbage keys to the download API.

3. Test coverage -- comprehensive: text+image, multiple images, no images, invalid JSON, empty content, title handling. All edge cases I can think of are covered.

4. Minimal change to extractContent -- replacing raw JSON passthrough with extractPostText() is the right call. The LLM gets clean text, and images are routed through the existing downloadInboundMedia pipeline.

5. appendMediaTags update -- the MsgTypePost addition to the image tag case is a simple one-liner that correctly reuses existing logic.

LGTM.

[yinwm]

Review Summary

This PR fixes a real issue where images embedded in Feishu post (rich text) messages were silently ignored.

What I like:

• Clean implementation: The new extractPostImageKeys() and extractPostText() functions follow the existing code style
• Good test coverage: Edge cases are well covered (text+image, multiple images, only images, invalid JSON, empty content)
• Reasonable fallback: extractPostText returns raw content on JSON parse failure, no message loss
• No breaking changes: The fix is additive and doesn't affect other message types

Minor observation (not blocking):

The same rawContent is JSON-parsed twice (once in extractPostImageKeys, once in extractPostText). This is a micro-optimization opportunity but acceptable for the current scale.

Suggestion:

Please complete the manual tests in the test plan before merging:

• Manual test: send image-only message in Feishu → agent sees image (regression check)
• Manual test: send text+image message in Feishu → agent now sees both text and image

Approve ✅

[yinwm]

fix linter plz, @Vast-Stars

[sipeed-bot]

@Vast-Stars Hi! This PR has had no activity for over 2 weeks, so I'm closing it for now to keep things tidy. If it's still relevant, feel free to reopen it anytime and we'll pick it back up.