Synchronize the main branch code to the office branch - #569
Merged
Merged
Conversation
configureDraftModel always labelled its exit choice "[Done]", which read as "finished" even when the provider and preset flows continue into endpoint, protocol, and capability prompts afterwards. Add an optional doneLabel option (defaulting to "[Done]") and pass "[Continue]" from the three call sites that have follow-up stages: provider configuration, new preset creation, and preset editing. Flows that genuinely end at the field menu (fallback models, channel, general settings) keep "[Done]". Both prompt test harnesses mapped the "done" fixture to the literal "[Done]"; they now resolve whichever exit label the current menu offers. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
重试成功后自动清除重试提示，避免在前端一直显示。 Changes: - 修改 finishRetryWaitStatusForTurn 从标记停止改为直接清除 - 更新相关测试用例验证状态被完全清除而非保留 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
L3 world model reaches the model only via the session L3 context (system-prompt) path, so turns/start no longer retrieves the L3 layer: - add turnStartMemoryLayers() and use it in SessionTurnService.startTurn and MemoryService.startTurnNoWrite so L3 is stripped even when requested - turnStartRetrievalLimit() sums tier1TopK + tier2TopK only (default 8); memory.search keeps the summed three-tier default (10) - drop the now-redundant strict-v2 L3 post-filter in RetrievalService.search - add tests covering layer stripping, read-only turn start, and the limit Co-authored-by: Cursor <cursoragent@cursor.com>
Official API ID deepseek-flash is native multimodal with 1M/384K limits; also register the informal deepseek-v4.1-flash ID so image input and implicit thinking match exact catalog lookups. Co-authored-by: Cursor <cursoragent@cursor.com>
episodes.title and episodes.summary were never written, so the task panel fell back to the first user and assistant messages. Add an episode_title worker job that fills both: a provisional pass after the first turn, a final pass on episode close. Input comes from raw_turns, because a rejected capture decision soft-deletes its L1 and leaves many episodes without one.
Codex stored auto-review prompts and injected context as user turns, so the task panel titled them as real conversations. Drop those turns in the shared rollout parser, and pin episode titles to the desktop language instead of guessing from the transcript.
Extend the Codex completeSourceTurn path to Cursor, Claude Code, OpenClaw, OpenCode, Hermes, and DeepSeek Harness so both channels share turn identity, tool pairing, and session binding.
fix dsh plugin error
fix false failure is reported when scanning meets abnormal data
optimize the display of scanning progress
Move PR-465 from 1.1.8 to 1.1.9
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
fix(memory): ignore XML metadata in import titles
…tion Fix/skill embedding input truncation
fix(memory): repair resume integrations across agents
The upstream v1.1.9 base already excludes the Literature Review cross-repository test harness and its npm commands. Co-authored-by: Cursor <cursoragent@cursor.com>
fix(models): prune stale assignments after partial deletion
Co-authored-by: Cursor <cursoragent@cursor.com>
…restart fix(models): preserve custom session model after restart
Kimi K3 fixes temperature=1.0 and rejects any explicit temperature value with HTTP 400. Add kimi-k3 to the immutable-temperature model list so requests omit the parameter. Co-authored-by: Cursor <cursoragent@cursor.com>
fix(providers): treat kimi-k3 as immutable-temperature model
In turn_start retrieval, pass up to the last 5 succeeded Q/A pairs of the current session (from the already-fetched recent RawTurns) to the retrieval.query.extract LLM call so it can resolve references such as "那个脚本" / "继续上次的" when producing queryVecText and keywords. - config: add algorithm.retrieval.queryExtractHistoryTurns (5) and queryExtractHistoryTextChars (200); 0 turns disables the feature - plugin-algorithms: RETRIEVAL_QUERY_EXTRACT_PROMPT v2 -> v3 with CURRENT USER INPUT / RECENT CONVERSATION rules and two few-shots - retrieval-service: export queryExtractHistoryFromRawTurns, render RECENT CONVERSATION + CURRENT USER INPUT for the extract user message, skip history when the current input exceeds 2000 chars - tests: bump extract operation to v3, add history selection, e2e, empty-history, non-turn_start, disabled and long-input cases, and config parsing coverage Only the extract model input changes; request protocol, candidate pool, LLM filter input, injection format and storage are untouched. Co-authored-by: Cursor <cursoragent@cursor.com> (cherry picked from commit c487f9b)
Unchanged turn re-imports were overwriting completed summaries with the queued placeholder and leaving processing marked ready. Co-authored-by: Cursor <cursoragent@cursor.com>
…ion_optimization-main
Claude/stream prefix dedup main
…ion_optimization-main
…tion-main Cursor/mem injection optimization main
fix(desktop): handle broken console output pipes
…titles fix(session): strip think blocks from generated conversation titles
fix: MEMMY-607 restore account after upgrade
test: MEMMY-594 preserve reinstall session semantics
fix: MEMMY-489 stop retrying exhausted quota
fix: MEMMY-487 recover interrupted upgrade locks
Co-authored-by: Cursor <cursoragent@cursor.com> # Conflicts: # Memory/src/service/memory-service.ts # Memory/tests/service/retrieval/query-and-filter.test.ts
Feature/agent wy
…sion The logout case uses the same empty active session as an upgrade, and the sync path already reactivates it from the account database. Co-authored-by: Cursor <cursoragent@cursor.com>
test: expect authoritative migration to restore a cleared account ses…
fix: keep integration service errors visible (MEMMY-623)
chore(release): merge v1.1.9 into main
syzsunshine219
requested review from
Wang-Daoji,
ZongYue99 and
hijzy
as code owners
September 29, 2026 07:05
syzsunshine219
merged commit Sep 29, 2026
fdb9dc9
into
archive/release-v1.1.6-with-office
22 checks passed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
6 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Synchronize the main branch code to the office branch