wangxuming/makelore
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

feat(agents): 完成云智能体工作台与交互流程

Browse Source
This commit is contained in:
brother7
2026-09-10 19:25:12 +08:00
parent 46fbae9dec
commit 64a3cf4934
26 changed files with 1955 additions and 49 deletions
Download Patch File Download Diff File Expand all files Collapse all files

47
.project-docs/30-worklog/tasks/20260910-personal-agent-platform-d15b2559.md Normal file
View File

@@ -0,0 +1,47 @@
# Task: Complete personal cloud Agent platform across WS Yuxi and MakeLore
## Identity
- Task ID: 20260910-personal-agent-platform-d15b2559
- Mode: Feature
- Branch: codex/20260910-personal-agent-platform-d15b2559-personal-agent-platform
- Worktree: D:\Datas\OthersProjects\.codex-worktrees\makelore\20260910-personal-agent-platform-d15b2559
- Base commit: 46fbae9dec70ee1bbfae671d458a81fd1d8a1c28
- Owner: codex
- Status: Ready for Integration
## Scope
- Complete the native cloud Agent workspace: configuration/preview, conversations/files/approval, publication/user sharing/application API, schedules/activity/cost; all cloud traffic and credentials remain Main-owned.
- User explicitly requested all remaining first-release modules, not another draft-only slice. Build on the reviewed entry commit recorded above; verify all accepted personal/share/API/scheduled/goal-tool paths before claiming completion.
## Intent And Constraints
- Bundled Concurrent Task Gate and Planning Gate passed. All operations are rooted in this owned isolated feature worktree. Peer records were read, with only the completed entry task added since the previous assessment; unrelated and defective peers remain read-only.
- Personal creators only; no organization/department/team wallet or Xiaozhi integration. WS owns identity/module/model eligibility and final Token Points; Yuxi owns cloud Request/Run/scheduler/resources. Existing WS Code/Canvas Gateway continues its own scope; this is the user-approved cloud-module extension recorded in the three-end plan.
- Creator, caller and payer remain distinct. Shared/API callers keep private conversations/files; every real model call charges the creator via existing WS accounting, never falls back to charging the caller. Application subjects are not fake personal accounts.
- Reuse Yuxi configuration, repositories, durable FIFO/Run/worker/approval, workspace and scheduling services; explicit empty capability selection disables it. Publish immutable configurations and freeze accepted requests, including resume/subcalls. Main owns cloud credentials/network, Renderer consumes safe Host API projections.
- Plan: verify and close model delegation/application scope contracts; implement server persistence/execution/finance and real transaction/worker tests; implement all native client surfaces and relevant Electron flows; run repository gates and document honest validation limits. No production data operations, deployment, push or additional subagents are implied.
## Outcome
- Completed native Agents workspace: My/Received/Activity, configuration and preview, formal conversations, approval/questions, attachments and downloadable artifacts, personal KB documents/indexing, publication/share/application credentials, creator costs and scheduled goals/history.
- Main owns exchanged credentials, actual cloud transport/SSE, account cancellation, local file dialogs/writes and desktop notifications; Renderer uses typed Host operations. Unsaved/unsent input and uncertain requests are preserved through explicit leave/retry flows; niancode Agent links route through login and server authorization.
- Implementation and verification are complete. The user explicitly authorized one fresh independent read-only Reviewer and commits after fixes. The fresh Reviewer reviewed the full three-repository scope and returned final PASS after confirming all findings were fixed. No unresolved review finding remains; no production deployment, push or merge was performed.
## Verification
- Passed: 32 tests across Main, draft/UI workflows and deep links. Includes stable retry intent, native batched SSE, approval restore, schedule enable intent, cost-service partial failure, actual temporary file bytes through mocked native dialogs, account switch cancellation and notification deduplication.
- Review fixes verified: new conversations appear immediately in history; unsent preview messages survive configuration edits; conversation changes protect approval answers and temporary attachments; cancelling file selection creates no empty conversation. The final attachment case verifies that only confirmation creates the conversation and associates the file. Renderer typecheck, targeted ESLint and Vite build passed again after this last fix.
- Passed: Renderer TypeScript; Vite build for Renderer/Main/preload/utility; real Electron Playwright scenario creates/saves/publishes an Agent, shows formal Markdown chat and creates an enabled schedule. The Host backend is explicitly stubbed in Electron coverage.
- Passed: targeted ESLint with no errors (2 non-blocking CloudChat warnings). Main TypeScript has the same 61 pre-existing errors, none in cloud Agents; this does not represent a clean repository-wide Main typecheck.
- Electron screenshots were inspected. No packaged installation, OS protocol registration or live three-service integration was executed.
## Follow-ups
- Hand off the reviewed result as committed feature branches for a separately authorized integration/release. Keep the worktrees and branches; no push, merge, deployment or further Reviewer is implied.
- Release-environment follow-up: actual one-api/paid-model reconciliation, provisioner sandbox execution, installable client packaging and production migration/rollback exercise. Current local validation does not claim these deployment outcomes.
## Promotion Candidates
- Target: integrated state and owning Agent/auth/billing/runtime architecture records. Proposal: complete the user-approved personal cloud platform using existing Yuxi capability/Request/Run/scheduling and WS accounting owners; document the cloud module's scope alongside the existing WS Gateway. Evidence: this task's final source and meaningful HTTP/PG/worker/Electron validation. Future impact: future modules consume one runtime and one ledger. Semantic conflicts: earlier planning fields and draft-only scope are superseded by implementation; existing unrelated product contracts remain. Human confirmation: product direction is already approved; canonical promotion remains a serialized integration task.