fix(feishu): include replied media context

[jukzthebest]

Summary

• Fetch normalized context for replied-to Feishu messages, including downloadable media resources.
• Attach replied parent media to inbound events so text replies to images/files keep the original attachment available to the agent.
• Keep existing reply-to text behavior through the new context helper and add regression coverage for parent image replies.

Test Plan

• python -m py_compile gateway/platforms/feishu.py tests/gateway/test_feishu.py
• python -m pytest tests/gateway/test_feishu.py -q

[alt-glitch]

Duplicate of #13115 (fix(feishu): include parent media context for replies). Also related to #24285, #26097, #26063 — all competing PRs for the same Feishu reply-to-media fix. Closes #26037.

[teknium1]

Thanks for the Feishu fix — I verified the bug still exists on current main, and the overall direction is useful.

Problems

• Current main only fetches parent text in _process_inbound_message (gateway/platforms/feishu.py:3132) and passes only the current message’s media_urls into MessageEvent (gateway/platforms/feishu.py:3171), so the premise is real.
• The patch adds _message_context_cache as a plain Dict on PR head (gateway/platforms/feishu.py:1453) and caches entries without eviction (gateway/platforms/feishu.py:3934). Current main already bounds reply text lookups with OrderedDict plus _FEISHU_MESSAGE_TEXT_CACHE_SIZE eviction (gateway/platforms/feishu.py:1458, gateway/platforms/feishu.py:4050). Salvaging this should preserve that bounded-cache behavior.

Suggested changes

• Use an OrderedDict/LRU cap for the new parent context cache, including move_to_end on hits and popitem(last=False) on overflow.
• Add coverage for parent file/audio media_refs and cache hit/eviction behavior; the downloader has separate image and media-ref paths (gateway/platforms/feishu.py:3643, gateway/platforms/feishu.py:3652).

Automated hermes-sweeper review.

[teknium1]

Thanks for the Feishu reply-context fix. The underlying bug is still present on current main: plugins/platforms/feishu/adapter.py:3259 fetches only parent text, while MessageEvent receives only the current inbound media at :3298-3299.

Problems

• The new _message_context_cache in this diff is an unbounded plain Dict. It retains downloaded parent-media paths indefinitely. Current main's analogous _message_text_cache is LRU-bounded at plugins/platforms/feishu/adapter.py:4165-4189; the salvaged context cache should preserve that behavior.
• The added test covers a parent image only. The downloader separately handles image keys and generic media references at plugins/platforms/feishu/adapter.py:3780-3799, so parent file/audio cases need coverage too.
• The adapter was relocated to plugins/platforms/feishu/adapter.py by 5600105478ffde29d7566b45421b100eaa29c4ef, so this needs a targeted salvage port rather than applying the old-path diff directly.

Suggested changes

• Port the helper and media merge to the bundled plugin, use a capped OrderedDict/LRU cache, and test image plus file/audio parent media and eviction behavior.

Automated hermes-sweeper review.

[teknium1]

Please make this a bounded OrderedDict/LRU cache. Parent contexts include downloaded media paths, so an unbounded Dict can grow for the lifetime of a busy adapter; current main's analogous reply-text cache promotes hits and evicts past _FEISHU_MESSAGE_TEXT_CACHE_SIZE.