Add notification digests with ON CONFLICT DO UPDATE, migrations 00068-00070,
supplementimpact module (analyze → confirm → claim-aware split), and wire
coordinator/mutate hooks plus proposal_required digests.
Anchor weather and other realtime questions on runner /v1/search results (or an explicit refuse constraint) so the model cannot invent dates or temperatures from training memory.
When summarizing the latest upload, poll parse state up to MinerU timeout,
emit document_parsing phase events so the bubble stays live and input stays
locked, then summarize or return a clear slow/failed reply.
Recognize “刚刚上传” as @latest_upload and summarize the project’s newest CAS attachment (with clear parse pending/failed/no-asset replies) instead of searching only controlled workflow docs.
Introduce internal/contracts/vocabulary for state/role/error sentinels,
alias through docgov, and retarget hot production consumers so
coordinator no longer imports docgov for vocabulary-only use.
Exact out-of-workflow documents can be summarized when WORKFLOW_SCOPE_MODE=weighted and the caller can still view the active instance; writes stay blocked.
Co-authored-by: Cursor <cursoragent@cursor.com>
Ask via PendingExpectation for weak-write verbs, mid confidence, or
stage-mismatched doc tasks instead of running blindly.
Co-authored-by: Cursor <cursoragent@cursor.com>
- ContextSnapshot gains ActiveWorkflowKey; ToSnapshot passes it downstream
- manifest-backed doc-title -> workflow index (main/pm/tm/rdev/sw)
- TRIGGER_TASK and DOCGOV write actions blocked with switch guidance when
the named document belongs to another active workflow
- legacy projects without an active workflow key keep the old behavior
Co-authored-by: Cursor <cursoragent@cursor.com>
Ambiguous consult only redirects on low confidence, and the testing template greets the signed-in user instead of a hardcoded name. Session tokens no longer parse as user id 0. Live chat sessions now evict the oldest entry at a hard cap.
Co-authored-by: Cursor <cursoragent@cursor.com>
Composition roots already hold Snapshot ports; the interface and archtest now forbid putting Pool/DocGov/DocRevision/ContextManager/CAS/ProjectWorkflow/FreeformDoc/KB back on Service.
Co-authored-by: Cursor <cursoragent@cursor.com>
Coordinator now takes a StageTaskRunner and injected context/KB ports.
Chatflow assembler reads maturity tokens instead of the orchestration root.
Co-authored-by: Cursor <cursoragent@cursor.com>
Move construction and document OnOutbox out of server, give HTTP/CLI a shared task use case, and block already-cleared import/event arch rules.
Co-authored-by: Cursor <cursoragent@cursor.com>
Sync coordinator and async CompletionSummary use pending_author_confirmation
with artifact metadata for document entry; completed only after commit.
Co-authored-by: Cursor <cursoragent@cursor.com>
Reactivate dev seed users after auth PG tests, use unique session tokens, fix import guard and coordinator assertions, and skip flaky E2E in short mode; add pg_restore CI drill job.
Co-authored-by: Cursor <cursoragent@cursor.com>
Add RUN_STATUS_QUERY/RUN_RESUME/RUN_RESTART with matcher and coordinator
handlers; write QUERY_PROGRESS/SUPPLEMENT_INFO subtypes to V2 frame; support
safe query+query multi-intent plans via IntentPlan dual-write.
Co-authored-by: Cursor <cursoragent@cursor.com>
- untrack internal/coordinator/dialogue_llm.go (kept on disk untracked as local WIP for a future dialogue-LLM commit with its own tests; it depends on engine.go fields (dialogueBackend, ReplySource) not yet committed)
- slim internal/chat/session.go back to Task 2 scope relative to bfa6889: keep EmitErrorInfo(stream.ClassifyError(...)) for intent+coordination errors and the dialogue_generating EmitPhase before ProcessIntent; drop sync.Mutex/mu, ChatResponse.ReplySource, preference/RecentDialogue injection block, and collectRecentDialogue helper (unrelated WIP for the dialogue-LLM feature)
Co-authored-by: Cursor <cursoragent@cursor.com>
weichen