Pin backend to the final-turn result checks #138
@@ -0,0 +1,70 @@
|
||||
# feature/agent-harness-p2 任务卡
|
||||
|
||||
任务主记录:https://git-home.u1680832.nyat.app:30319/weichen/Agent-System-Dev/pulls/138 ;状态:进行中;owner:weichen
|
||||
|
||||
## 基线与授权
|
||||
|
||||
- 权威仓:ssh://git@git-ssh.u1680832.nyat.app:31411/weichen/Agent-System-Dev.git;目标分支:main;R0(initial_base_sha)= d02fa9c20a594da287983d8edd47156ddbd49e01
|
||||
- 工作区:/home/weichen/projects/Agent-System-Dev-wt-feature-agent-harness-p2(worktree);分支:feature/agent-harness-p2
|
||||
- 授权与交付终点:实现 P2 结果把控。允许改 Agent-System-Backend,并在 `feature/agent-harness-p2` 上提 PR;Dev 仓只推进 `modules/backend` 并提 PR。未授权:合入 main、改 agent-native 内核、改小队路径、改 SubLLMRunner、做工作流执行器。
|
||||
- 启动自检(2026-10-01T13:44Z):`make precheck` verdict READY。head=d02fa9c20a,origin/main 同步,树干净,linked worktrees=0。必读 core/baseline.md、workflows/delivery.md §1 已读。
|
||||
|
||||
## 目标与非目标
|
||||
|
||||
- 目标:直连终轮在 `deliverableGate` 之后做确定性检查(工件存在且非空;md/txt/docx 走 quality,lint error 或评分不过即不通过),再由不带工具的验证器输出 `{passed, reason, nextAction}`。出错或无法解析判不通过并给 Notice。计划和待办连续 3 个工具轮未更新时提醒一次,每 run 最多 3 次,不落库。
|
||||
- 非目标:不改 agent-native;不改小队路径和 SubLLMRunner;不做工作流执行器。`runtime.verify` 缺省读取为 false,生产默认 true。
|
||||
|
||||
## 同步记录
|
||||
|
||||
| 节点 | sync_base_sha | tested_head_sha | checked_at |
|
||||
|---|---|---|---|
|
||||
| 启动 | d02fa9c20a594da287983d8edd47156ddbd49e01 | d02fa9c20a594da287983d8edd47156ddbd49e01 | 2026-10-01T13:45:07Z |
|
||||
| 同步前验收 | d02fa9c20a594da287983d8edd47156ddbd49e01 | f9d7bbe9f2cb4ccc3966faf3d3d7c2949dffcbc6 | 2026-10-01T13:54:53Z |
|
||||
|
||||
## 结果与验证
|
||||
|
||||
- 结果:直连终轮在 `deliverableGate` 之后检查本轮工件(存在、非空;md/txt/docx 走 `quality.ScoreBytes`),再由不带工具的验证器给出 `{passed, reason, nextAction}`。验证器出错或无法解析时不再续跑主模型,并附 Notice「本轮回复未通过独立验证(验证器不可用)」。计划和待办连续 3 个工具轮未更新提醒一次,每 run 最多 3 次,不落库。Backend `a3ae1b9`。Dev 只推进 `modules/backend` 到该提交。三处 agent-native 仍为 `a34689c`(Dev `deps/agent-native`、backend `deps/agent-native`、agent-squad `deps/agent-native`)。
|
||||
- 验证:见下方三段输出。`make test` 退出码 0(boundaries、core、backend、frontend vitest 51、ui vitest 71、agent-command)。
|
||||
- 影响与剩余项:`runtime.verify` 默认开启,每个终轮多一次轻量模型调用,可用该设置关闭。普通问答也会被「拿不准判不通过」,最多续跑 3 次。Backend PR https://git-home.u1680832.nyat.app:30319/weichen/Agent-System-Backend/pulls/44 与 Dev PR https://git-home.u1680832.nyat.app:30319/weichen/Agent-System-Dev/pulls/138 均未合入。小队路径与 SubLLMRunner 未改。
|
||||
|
||||
### precheck
|
||||
|
||||
```
|
||||
[repo] branch=main head=d02fa9c20a
|
||||
[remote] origin/main=d02fa9c20a ahead=0 behind=0
|
||||
[tree] clean submodules_drift=0 uninit=0
|
||||
[worktrees] linked=0
|
||||
[recycle] merged un-recycled branches: feature/card-close-20261001-124542Z feature/card-close-20261001-134217Z feature/card-close-base-extension-roadmap
|
||||
[required-reading] core/baseline.md, workflows/delivery.md §1
|
||||
[R0] d02fa9c20a(完整:d02fa9c20a594da287983d8edd47156ddbd49e01)
|
||||
[verdict] READY — 可建任务工作区:make task-start NAME=feature/your-branch
|
||||
```
|
||||
|
||||
### task-sync
|
||||
|
||||
```
|
||||
Already up to date.
|
||||
[synced] sync_base=d02fa9c20a tested_head=f9d7bbe9f2 checked_at=2026-10-01T13:54:53Z
|
||||
[next] 跑正式验收(make test 等),证据写入任务卡
|
||||
```
|
||||
|
||||
### make test
|
||||
|
||||
```
|
||||
bash scripts/lint-boundaries.sh
|
||||
boundaries ok
|
||||
cd modules/core && go test ./...
|
||||
ok agent-system-core 0.002s
|
||||
cd modules/backend && go test ./...
|
||||
(各包 ok,含 internal/runtime、internal/quality、internal/server)
|
||||
cd modules/frontend && npm ci && npx vitest run
|
||||
Test Files 9 passed (9)
|
||||
Tests 51 passed (51)
|
||||
cd modules/ui && npm ci && npx tsc --noEmit && npx vitest run
|
||||
Test Files 11 passed (11)
|
||||
Tests 71 passed (71)
|
||||
cd deps/agent-command && go vet ./... && go test -count=1 ./...
|
||||
ok agent-command 0.002s
|
||||
ok agent-command/conformance 0.002s
|
||||
```
|
||||
| 同步前验收 | d02fa9c20a594da287983d8edd47156ddbd49e01 | 9df951323ab83e6d3338b67c64e69156d3734989 | 2026-10-01T14:00:37Z |
|
||||
+1
-1
Submodule modules/backend updated: cfcf95d335...a3ae1b986d
Reference in New Issue
Block a user