Repository navigation
Conversation
…ent_core_develop # Conflicts: # sdk/nexent/core/agents/run_agent.py
Route Jiuwen model requests through the shared ModelService with trusted outlet snapshots, native tool calling and a controlled text protocol. Bridge authorized tools, MCP, Skills, subagents and A2A; preserve saved Core configuration and reuse Nexent context management and database history without replaying old tasks. Normalize public step observations and run results into existing Nexent chat events, including shared capacity metadata when context management is disabled. Add the v2.7.2 outlet migration, integration documentation and regression assets. Validation: previously completed Python/frontend UT and targeted formal checks; capacity regression rerun: 11 passed. Eight real-model business turns covered image analysis, follow-ups and Core switching. Formal real UI case remains failed due to AIDP 502; global asset gate remains blocked by F-111-RULE-02.
Merge develop at 5d94f69, retaining upstream service extraction, model controls and chat restoration fixes. Move Jiuwen persistence and output handling into the new execution/stream modules, retain trusted outlet provisioning and model snapshots in the refactored record factory, and adapt regression fixtures to the new module boundaries. Synchronize the catalog content digest after prior whitespace normalization. Validation: 2139 distinct Python UT, 52 frontend tests, TypeScript and 3 formal Cases (28 assertions) passed. Global formal asset validation remains blocked by existing F-111-RULE-02. No fresh deployment or real-model D4 performed.
Propagate saved Skill authorization snapshots, provide controlled native Python file generation, and resolve registered MCP names to original wire names while rejecting collisions. Add unit regressions and deployed resource/Skill journey cases covering files, knowledge retrieval, external MCP, mixed-core delegation, and follow-up history without replay.
Adapt Jiuwen preparation, model outlet snapshots and conversation ordering to the canonical backend structure. Preserve structured clarification responses and immutable historical SQL migrations.
Include d954d64 and refresh formal asset views. Supply the missing scrollIntoView browser boundary in the citation test and assert selected-source scrolling.
Separate contracts, resource orchestration, Core context bridges and sandbox modules. Migrate SDK, backend and test import sites and remove legacy alias modules. Preserve model implementations and backend business behavior. Validation: 92 isolated contract and import-boundary tests; syntax, copyright, unchanged backend business AST and frozen model blobs checked.
Centralize history, Skills, artifact and response rules across smolagents and Jiuwen. Share current-request delegation intent, authorized targets, successful dispatch state and bounded recovery while retaining Core-specific execution protocols. Add policy and protocol regressions and a two-Core file-assistant browser journey. Validation of the exact commit tree: 234 Python regressions and 14 controlled runtime tests. Previous delivery also passed 739 Python tests, 14 frontend tests and both real-model browser variants; retain the separately recorded global asset-gate limitation.
Use canonical context/configuration imports in backend, benchmarks and isolated regression stubs. Align sandbox and context tests with the already migrated SDK, retain terminal MCP error guards, and include generated navigation for the Jiuwen feature contracts. Remove duplicate copyright lines without changing model modules. Validation: isolated SDK suites 1,616 passed; terminal guard 5 passed; benchmark CLI imports, copyright and formal asset checks passed.
Export and import the saved agent core and context-manager enable flag. Keep smolagents and enabled CM as backward-compatible defaults for older payloads, and reject unsupported cores through the public contract. Validation: 12 focused transfer tests and formal asset checks passed.
Update fixture authentication and unified prompt fields, include the controlled Python tool in native-mode expectations, and publish owned fixtures before checking their exports. Support explicit browser channel selection and distinguish resource diagnostic subsets from full passes. Include the existing delegation fix traceability record. Validation: model/native-file UT 71 passed; model bridge D3 4 passed; resource D3 6 passed and D5 4 passed; formal asset checks passed.
Authorize explicit saved-user retry targets and acknowledge their server identity without appending a new question. Keep history siblings ordered, project the latest answer into model context, and share running-branch recovery across chat and Workbench without creating an extra branch. Include required retry helpers, boundary regressions and fixed browser acceptance assets. Independently repeated equal-text questions retain their own persisted boundaries. Validation: backend UT 34 and frontend UT 35 passed; type-check passed; UT-FE-040 passed all 5 component tests; formal asset checks passed. Real browser rerun was blocked by model_authentication_error on the first model call. Owned fixtures were cleaned and the source stayed unchanged; this rerun is not a real-model acceptance pass.
Merge develop a73ab25 while retaining canonical SDK contracts, factory, orchestration, resources, output and per-Core implementation directories. Port upstream execution-history, callable-name and MCP diagnostics changes to their canonical owners; do not restore removed SDK aliases. Keep shared prompt/delegation rules, explicit output results and retry history branches. Preserve Qwen tool ID continuations alongside updated stream trailers and provider diagnostics. Freeze Jiuwen connection headers and use canonical model outlet columns ahead of transitional snapshots. Validate 2,922 Windows unit tests, 2 Linux symlink checks, 92 frontend tests, frontend types, 8 selected formal cases and combined-tree asset gates. Local configuration, test results and unrelated untracked files excluded.
Use the deployment-wide Core policy and remove per-Agent Core persistence and the unmerged selection migration. Preserve Agent models, CM and resources. Keep Jiuwen executions on a managed event loop, bridge terminal ask_user forms, normalize legacy model preferences and preserve native history units. Restore MCP creator interactions and waiting terminals, retain tool identities and Skill artifacts in history, and show generated-Agent result cards only on matching completion events. Keep draft editing as a separate action. Include regression tests, formal Cases and integration documentation. Validation: 278 targeted Python tests, 104 frontend tests, copyright and whitespace checks, and strict design/implementation asset validation passed. Real Workbench journeys on both Cores passed on the matching deployed code.
Compare exact sorted Skill-name lists when checking the owned clone. Retain duplicate detection and the existing model, prompt and tool checks. Record the test-only correction with its owning formal Case.
Merge develop at 5876d6e. Keep canonical SDK contracts, public capabilities and per-Core implementations; do not restore legacy aliases or move models. Apply upstream creation-confirmation validation and literal clarification wrappers in the shared contract. Retain smolagents cancellation, bounded clarification recovery and actionable Python syntax feedback. Omit absent optional clarification fields from Jiuwen events so the shared frontend continues to render ordinary ask_user cards. Migrate new upstream test imports and align duplicate-usage assertions with final snapshot emission. Validation: 2429 Windows Python tests and two Linux symlink tests passed; 185 Workbench and 119 component tests passed. Two component timeouts were reproduced against origin/develop sources and remain separately recorded. Three D3 Cases passed 83 variants with cleanup. Strict asset validation, TypeScript, copyright and whitespace checks passed. No deployment or push.
Add a real Workbench ask_user journey covering terminal forms, exact saved response identity, read-only history, follow-up facts and owned cleanup. Align the Agent creator regression with the Submitted history label and verify restored text, selected/unselected options and other answers against the actual submitted payload. Register the Case, binding and change record. Validation: 161 existing Python regressions and three real-model Workbench D4 journeys passed. Selected files match the verified source hashes. Strict design/implementation asset validation, copyright and whitespace checks passed.
YehongPan
requested review from
Dallas98,
Jasonxia007,
WMC001,
hhhhsc701 and
jeffwu-1999
as code owners
November 10, 2026 07:26
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
…utdown (#4316) * fix(deploy): recover removed Supabase source during upgrades * fix(deploy): stop legacy Config only for pending Workbench migration
This branch has not been deployed
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
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.
✨ Feature: Introduce jiuwen as an optional agent core.
[Specification Details]
factory、orchestration、resources、output、context、sandbox、cores。两种 Core 的专属实现分别归档，项目调用统一迁移到新路径。[Test Results]
人在回路
工具调用（图片分析）
子智能体调用
nl2agent适配
nl2skills适配
并行调用（多子智能体协同）
工作台知识库查询挂载
工作台智能体挂载
工作台技能挂载