Best possible solution:

Refresh this patch onto current main or open an equivalent mergeable replacement, keep the Feishu regression coverage and changelog entry, run pnpm check:changed, then land it and close #38806 while leaving #66631 and #70512 tracked separately.

Do we have a high-confidence way to reproduce the issue?

Yes. A source-level reproduction is a Feishu event with chat_type: "p2p", root_id, and thread_id: current main parses those fields, then routes the message with P2P replyInThread false and reply metadata skipped.

Is this the best way to solve the issue?

Yes, after branch refresh. The PR’s approach is the narrow maintainable fix because it uses P2P thread_id as the thread-reply opt-in while preserving root-id-only quote replies and existing group behavior.

What I checked:

• current-main-parses-thread-fields: Current main parses root_id and thread_id into the Feishu message context, so the reported P2P thread signal is available to routing code. (extensions/feishu/src/bot.ts:250, bd6035d97707)
• current-main-disables-p2p-thread-replies: Current main still forces replyInThread to false for non-group Feishu messages. (extensions/feishu/src/bot.ts:717, bd6035d97707)
• current-main-strips-p2p-reply-targets: Current main passes skipReplyToInMessages: !isGroup, which makes P2P dispatch drop replyToMessageId before the send layer can call the Feishu reply API. (extensions/feishu/src/bot.ts:1315, bd6035d97707)
• send-layer-supports-thread-reply: The Feishu send helper already calls client.im.message.reply and sends reply_in_thread: true when replyInThread is provided, so the missing behavior is routing P2P thread events into that existing path. (extensions/feishu/src/send.ts:167, bd6035d97707)
• pr-patch-routes-p2p-thread-events: The PR derives directThreadReply from P2P thread_id, targets the direct thread root, preserves reply metadata for those events, and sets thread-reply behavior. (extensions/feishu/src/bot.ts:714, 2f9535c226ca)
• pr-patch-adds-regression-tests: The PR adds regression coverage for P2P thread_id messages and for root-id-only P2P quote replies preserving the non-thread path. (extensions/feishu/src/bot.test.ts:2573, 2f9535c226ca)

Likely related people:

• guoqunabc: Merged PR fix(feishu): comprehensive reply mechanism — outbound replyToId forwarding + topic-aware reply targeting #33789 changed Feishu reply targeting around root/thread handling, which is the regression boundary identified by the linked bug and this PR. (role: introduced relevant reply-targeting behavior; confidence: high; commits: 63ce7c74bdc0; files: extensions/feishu/src/bot.ts, extensions/feishu/src/send.ts, extensions/feishu/src/outbound.ts)
• steipete: Recent current-main Feishu and shared channel refactors touched the affected bot, test, send, and reply-dispatcher surfaces. (role: recent maintainer; confidence: high; commits: 9a9cd0c0abdf, ffe67e9cdc9e, b388209eaf72; files: extensions/feishu/src/bot.ts, extensions/feishu/src/bot.test.ts, extensions/feishu/src/send.ts)
• vincentkoc: Recent Feishu work touched send/reply-dispatcher-adjacent code, and this PR discussion includes a maintainer automerge signal from this account. (role: recent adjacent maintainer; confidence: medium; commits: 2b96f53f9782, c6cf37068cae, 84a22a64bef0; files: extensions/feishu/src/bot.ts, extensions/feishu/src/send.ts, extensions/feishu/src/reply-dispatcher.ts)

Remaining risk / open question:

• The PR branch is dirty, non-rebaseable, and 1568 commits behind current main, so it cannot land as-is.
• pnpm check:changed was not run during this read-only review; pnpm is unavailable in this shell and validation should run after the branch is refreshed.

Codex review notes: model gpt-5.5, reasoning high; reviewed against bd6035d97707.