What's Changed
- test(managed-agent): observe a real scheduled /loop fire in the full-stack E2E by @wenshao in https://github.com/QwenLM/qwen-code/pull/17506
- fix(verify-pr): harden the watch-state shape guard, container nonce, and VM prose by @wenshao in https://github.com/QwenLM/qwen-code/pull/17440
- fix(external-context): deflake cumulative-deadline deletion timing by @qwen-code-dev-bot in https://github.com/QwenLM/qwen-code/pull/17342
- docs(design): propose idle Session suspension toward a stateless Hosted Harness by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/17527
- docs(design): drop the 20-run parity requirement from the loop rollout steps by @qqqys in https://github.com/QwenLM/qwen-code/pull/17535
- fix(serve): reserve listener headroom from SSE and reap stalled writers by default by @wenshao in https://github.com/QwenLM/qwen-code/pull/17502
- feat(web-shell): explain background tasks from the composer chips by @wenshao in https://github.com/QwenLM/qwen-code/pull/17398
- fix(triage): fail closed when QWEN_TRIAGE_DRAFT_DIR is unset in CI by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/17549
- test(channels): deflake the DingTalk webhook-cache recency case by @qwen-code-dev-bot in https://github.com/QwenLM/qwen-code/pull/17615
- fix(managed-agent): recover vanished Harness admissions safely by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/17521
- feat(web-shell): add the unified chat surface contract and capability guards by @wenshao in https://github.com/QwenLM/qwen-code/pull/17520
- feat(channels): add GitHub as the third managed channel adapter by @wenshao in https://github.com/QwenLM/qwen-code/pull/17511
- perf(core): bound follow-up suggestion context by @callmeYe in https://github.com/QwenLM/qwen-code/pull/17532
- feat(generated-ui): let cards save view state through the host by @qqqys in https://github.com/QwenLM/qwen-code/pull/17522
- feat(web-shell): send to another session with /send by @qqqys in https://github.com/QwenLM/qwen-code/pull/17539
- fix(web-shell): serve PR watch deep links before authentication by @Afloat16 in https://github.com/QwenLM/qwen-code/pull/17119
- feat(web-shell): decouple the shared composer from daemon value imports by @wenshao in https://github.com/QwenLM/qwen-code/pull/17661
- fix(core): converge Read unchanged-dedup with the preview lineage by @callmeYe in https://github.com/QwenLM/qwen-code/pull/17465
- fix(web-shell): bottom-sheet model picker on touch screens by @callmeYe in https://github.com/QwenLM/qwen-code/pull/17401
- feat(channels): add generic outbound webhook delivery by @jkxiao in https://github.com/QwenLM/qwen-code/pull/17610
- fix(core): scope Grep fallback strategies to the git-aware corpus by @callmeYe in https://github.com/QwenLM/qwen-code/pull/17530
- fix(managed-agent): drive continuation-ready parks in the epoch-less resubmission arm by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/17680
- fix(web-shell): align managed errors with DAC and stabilize recovery by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/17557
- fix(channels): keep DingTalk multi-chunk sends untruncated; the ledger converges by @wenshao in https://github.com/QwenLM/qwen-code/pull/17529
- feat(managed-agent): notify attachment retirement and age out suspension waits by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/17681
- fix(serve): Prevent automatic work from blocking worktree restore by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/17654
- fix(web-shell): keep each /send reply in view and drop the gap under send records by @qqqys in https://github.com/QwenLM/qwen-code/pull/17686
- fix(verify-pr): register the watch round ahead of prepare (R17-2 liveness) by @wenshao in https://github.com/QwenLM/qwen-code/pull/17550
- fix(core): release idle JSONL append locks by @Afloat16 in https://github.com/QwenLM/qwen-code/pull/17577
- fix(web-shell): retry memory content after refresh by @Afloat16 in https://github.com/QwenLM/qwen-code/pull/17108
- fix(sdk): apply caller headers without case-sensitive duplicates by @Afloat16 in https://github.com/QwenLM/qwen-code/pull/17579
- fix(cli): preserve pending cleanup when handlers unregister by @Afloat16 in https://github.com/QwenLM/qwen-code/pull/17572
- fix(sdk): stop buffered SSE delivery after cancellation by @Afloat16 in https://github.com/QwenLM/qwen-code/pull/17569
- fix(java): cancel heartbeats removed during registration by @Afloat16 in https://github.com/QwenLM/qwen-code/pull/17117
- fix(cli): preserve literal custom command arguments by @Afloat16 in https://github.com/QwenLM/qwen-code/pull/17367
- feat(web): configure generic webhook destinations by @jkxiao in https://github.com/QwenLM/qwen-code/pull/17405
- feat(web-shell): show the session artifact count at click time from list metadata by @wenshao in https://github.com/QwenLM/qwen-code/pull/17518
- feat(acp): run Mods in ACP sessions by @qqqys in https://github.com/QwenLM/qwen-code/pull/17693
- refactor(web-shell): extract the shared composer status stack component by @wenshao in https://github.com/QwenLM/qwen-code/pull/17684
- fix(sdk-python): isolate initialization from caller cancellation by @Afloat16 in https://github.com/QwenLM/qwen-code/pull/17499
- feat(serve): unify SSE admission and occupancy across every stream surface by @wenshao in https://github.com/QwenLM/qwen-code/pull/17662
- feat(channels): add WeCom adapter to the managed channel path by @wenshao in https://github.com/QwenLM/qwen-code/pull/17441
- fix(web-shell): make background work visible on sidebar session rows by @wenshao in https://github.com/QwenLM/qwen-code/pull/17676
- test(managed-agent): align read-retry versioning count with the probe removal by @wenshao in https://github.com/QwenLM/qwen-code/pull/17082