feat: add engineering evaluation review and adoption UI

This commit is contained in:
Agent Workbench Dev
2026-11-01 09:56:25 +08:00
parent 2a44f15693
commit 17c9f9d0cd
16 changed files with 491 additions and 123 deletions
+9 -1
View File
@@ -7,7 +7,7 @@
| A 固定契约 | 专项通过 | 1–3固定样本、HEAD/同源字节、明确源码/原测试/检查命令、模型和总预算 |
| B 运行账本 | 专项通过 | 原版0模型请求、候选原工具/审批/预算、停止与重启未知、不重放已尝试准备 |
| C 比较 | 专项通过 | 原失败/全部尝试/日志SHA、原用量缺失及人工审批/应用、不可变报告和父审查/Git门控 |
| D 产品流程 | 待实现 | 桌面/390px准备、审查、比较和采纳 |
| D 产品流程 | 7组通过 | 桌面/390px准备、审查、比较、导出、刷新、采纳、停止和来源变化;两视口实际截图回看 |
| E 实际工程 | 待验 | Node/React/Python及自身模块小范围试点 |
| F 完整发行 | 待验 | 完整源码/历史UI/独立源码/便携/安装/升级/原生回看 |
@@ -22,3 +22,11 @@
最后原版日志/采纳/HTTP3条路径3/3通过(`.qa/evaluation-abc-final-guards.log`)。首次选择器错误后还发现受控textarea的标签名称会随内容变化,现全部关键textarea均有稳定的aria-label,另开收据复验;原第二次失败`.qa/engineering-evals-ui-b08b6c81-cd2b-47c5-85d4-66ecba9fd922`保持。
父工程检查/变更集/Git/工作区各操作的活动门控已纳入评测活动,最终相关2路径2/2通过。A-C原日志按原字节封存于[清单](engineering-evals-abc-evidence/manifest.json)。
## D 产品流程回顾
最终7组通过,原收据`.qa/engineering-evals-ui-7203e02f-ee05-41ab-81dd-40b215086863`,概览见[实际UI记录](engineering-evals-ui-validation.json)。实际原版3测试失败、候选3测试通过、父审查/检查/Git均完成;停止保留stopped,父来源变化在创建工作区前拒绝。桌面并列和390px堆叠截图已目视回看,原失败与候选通过、人工审批/应用次数、缺失费用及固定响应标签均可读,未见页面横向溢出。
第三次UI收据`.qa/engineering-evals-ui-d240669e-0c9b-4450-960e-1b870f9fb23f`前5组已通过,之后QA在POST完成前读取最新批次,误选旧批次导致停止场景等待失败。修正为等待批次数量和新标题后获取原对象;原失败保持,最终轮独立执行完整7组。此为验收脚本竞态,不删除或改写原失败。
A-C原日志在提交时被Git默认换行规范化;现增加证据目录字节保护,并重新按原`.qa`字节暂存,后续提交逐Git blob核对清单SHA。保留旧提交,不改写历史。
+6
View File
@@ -17,3 +17,9 @@
复用了原WorkerManager/WorkspaceManager、CheckRunner、ChangeSet及GitDelivery。原版不伪装成一个有修改的worker成果,评测归集只接受候选;普通协作接口不能绕过评测开始或采纳。全部检查尝试、原日志SHA及上游未知用量进入固定报告,缺失证据显示unknown并阻止采纳,不删除原分母。多样本结束后只能采纳一项;先核对原父来源,再切回所属Task形成独立变更集,不能直接覆盖父文件。
回顾发现采纳预检含异步窗口,已增加父工程活动锁和持久采纳意图;重启不恢复副作用。计划停止也等待原准备结束。对源码/模型修订、精确文件、原测试、HTTP归属及停止恢复的12项专项和原worker6项合计18/18通过,原10项首轮及后续修正证据分别保留。工程推理成功率、真实模型费用和Purge效果没有据此判为完成。
## D 回顾
实际原任务导航、提案审查、检查审批、对照报告与采纳后父交付7组通过。对照面板将固定响应标签放在报告头部,390px按原版/候选顺序堆叠并实际滚动核对候选测试证据;缺失费用显示未报告。修正可访问名称随select选项/textarea内容变化的问题,关键表单有稳定名称。第三轮QA提前读取最新对象的竞态修正后,第四轮完整通过;原失败收据保持。复用了现有模态、WorkerPanel、差异审查和Git交付,没有新增执行内核或依赖。
回看补充同批切换样本时迟到报告响应的当前样本身份核对,避免旧报告盖住新选择;TypeScript通过,随后完整历史/新评测回归覆盖最终代码。证据目录保留原CRLF字节并显式声明cr-at-eol,避免Git文本规范化改变日志及清单。
@@ -1,18 +1,18 @@
✔ evaluation contract fixes bounded samples, source paths, checks and aggregate budget (525.0705ms)
✔ same-source workspaces seal real commands before execution and preserve original staged and unstaged bytes (11083.0145ms)
✔ actual original failures and candidate success are both preserved, then adoption requires parent review, checks and Git (40331.2376ms)
✔ candidate cannot shrink tests and changing original test bytes rejects execution before any model request (15626.3462ms)
✔ missing upstream usage remains unknown and does not become a free run or invented cost (25914.3911ms)
✔ missing raw check logs preserve unknown result and original sample denominator (25759.3229ms)
✔ source and model changes never silently replace an existing batch input (2304.4183ms)
✔ sample queue cannot skip or reset attempted samples, and reports freeze further writes (20862.9625ms)
✔ stop cancels only current candidate and preserves partial results instead of replaying a request (16535.6083ms)
✔ restart recovers interrupted evaluations as unknown and keeps completed checks and plan identity (15402.608ms)
ℹ tests 10
ℹ suites 0
ℹ pass 10
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 174485.0587
✔ evaluation contract fixes bounded samples, source paths, checks and aggregate budget (525.0705ms)
✔ same-source workspaces seal real commands before execution and preserve original staged and unstaged bytes (11083.0145ms)
✔ actual original failures and candidate success are both preserved, then adoption requires parent review, checks and Git (40331.2376ms)
✔ candidate cannot shrink tests and changing original test bytes rejects execution before any model request (15626.3462ms)
✔ missing upstream usage remains unknown and does not become a free run or invented cost (25914.3911ms)
✔ missing raw check logs preserve unknown result and original sample denominator (25759.3229ms)
✔ source and model changes never silently replace an existing batch input (2304.4183ms)
✔ sample queue cannot skip or reset attempted samples, and reports freeze further writes (20862.9625ms)
✔ stop cancels only current candidate and preserves partial results instead of replaying a request (16535.6083ms)
✔ restart recovers interrupted evaluations as unknown and keeps completed checks and plan identity (15402.608ms)
ℹ tests 10
ℹ suites 0
ℹ pass 10
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 174485.0587
@@ -1,11 +1,11 @@
✔ actual original failures and candidate success are both preserved, then adoption requires parent review, checks and Git (42230.2742ms)
✔ missing raw check logs preserve unknown result and original sample denominator (31180.7936ms)
✔ HTTP evaluations require host token, original fingerprint and session ownership; worker API cannot bypass evaluation flow (15068.7458ms)
ℹ tests 3
ℹ suites 0
ℹ pass 3
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 88729.2571
✔ actual original failures and candidate success are both preserved, then adoption requires parent review, checks and Git (42230.2742ms)
✔ missing raw check logs preserve unknown result and original sample denominator (31180.7936ms)
✔ HTTP evaluations require host token, original fingerprint and session ownership; worker API cannot bypass evaluation flow (15068.7458ms)
ℹ tests 3
ℹ suites 0
ℹ pass 3
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 88729.2571
@@ -1,10 +1,10 @@
✔ actual original failures and candidate success are both preserved, then adoption requires parent review, checks and Git (41949.8076ms)
✔ HTTP evaluations require host token, original fingerprint and session ownership; worker API cannot bypass evaluation flow (12432.7767ms)
ℹ tests 2
ℹ suites 0
ℹ pass 2
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 54628.6602
✔ actual original failures and candidate success are both preserved, then adoption requires parent review, checks and Git (41949.8076ms)
✔ HTTP evaluations require host token, original fingerprint and session ownership; worker API cannot bypass evaluation flow (12432.7767ms)
ℹ tests 2
ℹ suites 0
ℹ pass 2
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 54628.6602
@@ -1,26 +1,26 @@
✔ evaluation contract fixes bounded samples, source paths, checks and aggregate budget (879.7691ms)
✔ same-source workspaces seal real commands before execution and preserve original staged and unstaged bytes (14874.6757ms)
✔ actual original failures and candidate success are both preserved, then adoption requires parent review, checks and Git (49362.4978ms)
✔ candidate cannot shrink tests and changing original test bytes rejects execution before any model request (19490.2731ms)
✔ missing upstream usage remains unknown and does not become a free run or invented cost (28035.4922ms)
✔ missing raw check logs preserve unknown result and original sample denominator (27592.2318ms)
✔ source and model changes never silently replace an existing batch input (2375.6565ms)
✔ model revision change before preparing a fixed sample cannot silently use the replacement connection (2210.0987ms)
✔ HTTP evaluations require host token, original fingerprint and session ownership; worker API cannot bypass evaluation flow (12196.9645ms)
✔ sample queue cannot skip or reset attempted samples, and reports freeze further writes (21512.0204ms)
✔ stop cancels only current candidate and preserves partial results instead of replaying a request (17772.6846ms)
✔ restart recovers interrupted evaluations as unknown and keeps completed checks and plan identity (16171.5361ms)
✔ worker scopes and allocated budgets reject aliases, extra permissions and aggregate overflow (770.0996ms)
✔ two sequential Git workspaces freeze a new parent goal, separate owned tasks and preserve all source bytes (13643.8712ms)
✔ real isolated changes, approvals and npm checks produce original owned outputs, not model completion claims (28402.4151ms)
✔ forged worker tools cannot execute terminal or Git even when absent from the advertised tool list (12969.4039ms)
✔ a completion-only model pauses each worker without fabricated changes or checks (14462.927ms)
✔ restart retains frozen scope and marks interrupted parent and workers unknown without sending requests (12239.9837ms)
ℹ tests 18
ℹ suites 0
ℹ pass 18
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 212727.8273
✔ evaluation contract fixes bounded samples, source paths, checks and aggregate budget (879.7691ms)
✔ same-source workspaces seal real commands before execution and preserve original staged and unstaged bytes (14874.6757ms)
✔ actual original failures and candidate success are both preserved, then adoption requires parent review, checks and Git (49362.4978ms)
✔ candidate cannot shrink tests and changing original test bytes rejects execution before any model request (19490.2731ms)
✔ missing upstream usage remains unknown and does not become a free run or invented cost (28035.4922ms)
✔ missing raw check logs preserve unknown result and original sample denominator (27592.2318ms)
✔ source and model changes never silently replace an existing batch input (2375.6565ms)
✔ model revision change before preparing a fixed sample cannot silently use the replacement connection (2210.0987ms)
✔ HTTP evaluations require host token, original fingerprint and session ownership; worker API cannot bypass evaluation flow (12196.9645ms)
✔ sample queue cannot skip or reset attempted samples, and reports freeze further writes (21512.0204ms)
✔ stop cancels only current candidate and preserves partial results instead of replaying a request (17772.6846ms)
✔ restart recovers interrupted evaluations as unknown and keeps completed checks and plan identity (16171.5361ms)
✔ worker scopes and allocated budgets reject aliases, extra permissions and aggregate overflow (770.0996ms)
✔ two sequential Git workspaces freeze a new parent goal, separate owned tasks and preserve all source bytes (13643.8712ms)
✔ real isolated changes, approvals and npm checks produce original owned outputs, not model completion claims (28402.4151ms)
✔ forged worker tools cannot execute terminal or Git even when absent from the advertised tool list (12969.4039ms)
✔ a completion-only model pauses each worker without fabricated changes or checks (14462.927ms)
✔ restart retains frozen scope and marks interrupted parent and workers unknown without sending requests (12239.9837ms)
ℹ tests 18
ℹ suites 0
ℹ pass 18
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 212727.8273
@@ -1,48 +1,48 @@
{
"stage": "1.20 A-C",
"first": {
"passed": 10,
"failed": 0
},
"reviewed": {
"evaluations": 12,
"workers": 6,
"passed": 18,
"failed": 0
},
"finalGuardPaths": {
"passed": 3,
"failed": 0
},
"finalOperationLockPaths": {
"passed": 2,
"failed": 0
},
"realModelRequests": 0,
"files": [
{
"source": ".qa\\evaluation-a-first.log",
"file": "evaluation-a-first.tap",
"bytes": 1259,
"sha256": "8d5ba79071a3339a5ac2affdcb13a6af88f64fc9c4932c209bca53a0da6b3caf"
},
{
"source": ".qa\\evaluation-abc-reviewed.log",
"file": "evaluation-abc-reviewed.tap",
"bytes": 2228,
"sha256": "15b798551aa6dee8e5b12d73fc070272d1c67b6ffdbc6445d41fcb63d95ec8fa"
},
{
"source": ".qa\\evaluation-abc-final-guards.log",
"file": "evaluation-abc-final-guards.tap",
"bytes": 504,
"sha256": "2db5ee72d4439acb97176bedb14985ffe91cc7ecbc361e32a969b08b39a340f9"
},
{
"source": ".qa\\evaluation-abc-operation-locks.log",
"file": "evaluation-abc-operation-locks.tap",
"bytes": 405,
"sha256": "f689978a90d947c0561dc8f5a891b51b432b547dc206fe302279566957a6c75d"
}
]
}
{
"stage": "1.20 A-C",
"first": {
"passed": 10,
"failed": 0
},
"reviewed": {
"evaluations": 12,
"workers": 6,
"passed": 18,
"failed": 0
},
"finalGuardPaths": {
"passed": 3,
"failed": 0
},
"finalOperationLockPaths": {
"passed": 2,
"failed": 0
},
"realModelRequests": 0,
"files": [
{
"source": ".qa\\evaluation-a-first.log",
"file": "evaluation-a-first.tap",
"bytes": 1259,
"sha256": "8d5ba79071a3339a5ac2affdcb13a6af88f64fc9c4932c209bca53a0da6b3caf"
},
{
"source": ".qa\\evaluation-abc-reviewed.log",
"file": "evaluation-abc-reviewed.tap",
"bytes": 2228,
"sha256": "15b798551aa6dee8e5b12d73fc070272d1c67b6ffdbc6445d41fcb63d95ec8fa"
},
{
"source": ".qa\\evaluation-abc-final-guards.log",
"file": "evaluation-abc-final-guards.tap",
"bytes": 504,
"sha256": "2db5ee72d4439acb97176bedb14985ffe91cc7ecbc361e32a969b08b39a340f9"
},
{
"source": ".qa\\evaluation-abc-operation-locks.log",
"file": "evaluation-abc-operation-locks.tap",
"bytes": 405,
"sha256": "f689978a90d947c0561dc8f5a891b51b432b547dc206fe302279566957a6c75d"
}
]
}
+73
View File
@@ -0,0 +1,73 @@
{
"passed": true,
"version": "1.19.0",
"evidenceKind": "owned fixed response, actual isolated Git/checks and desktop/390px UI",
"realModelRequests": 0,
"fixtureRequests": 6,
"assertions": [
"desktop/390px form rejects candidate test edits, keeps entered requirements, persists fixed model/source/budget and reloads the original plan",
"explicit preparation seals two same-source checkouts and actual commands; original real three-test failure is retained with zero model requests",
"candidate navigates to its original restricted task for real change review and check approval, runs the original test/build and preserves the parent source",
"parallel comparison preserves original failure, candidate pass, real approval/application counts and absent cost; export and refresh retain the original report and fixture label",
"adoption forms a fresh parent review batch; completion requires applied source, fresh real parent checks and local Git delivery while owner modifications remain exact",
"stop waits for the owned candidate request, preserves stopped outcome and never turns cancellation into a passed sample or a replay",
"changed owner source is rejected before workspace creation and the error stays attached to the original batch instead of a generic chat reply"
],
"screenshots": [
{
"file": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\engineering-evals-ui-7203e02f-ee05-41ab-81dd-40b215086863\\desktop-empty.png",
"sha256": "a2180342db5c91cca02d41558019c334b83314eba89863005123813d68cce19e"
},
{
"file": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\engineering-evals-ui-7203e02f-ee05-41ab-81dd-40b215086863\\mobile-form-scope-error.png",
"sha256": "79ff7a3dd7a3e016cbc61ded4e41f7298085fbd0a7256ba0d5d6aa15c403f7ac"
},
{
"file": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\engineering-evals-ui-7203e02f-ee05-41ab-81dd-40b215086863\\desktop-ready.png",
"sha256": "743a33128dbf82a517bcda9f116b73c6b3ac8f0a800747c06561c60efc275774"
},
{
"file": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\engineering-evals-ui-7203e02f-ee05-41ab-81dd-40b215086863\\desktop-candidate-complete.png",
"sha256": "536cb42c8840112a2069682624428bcecb6ab79863b2296bf5542f89c86faffd"
},
{
"file": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\engineering-evals-ui-7203e02f-ee05-41ab-81dd-40b215086863\\desktop-comparison.png",
"sha256": "12089849e6fc1113946abcd24c5208a6413a7d8f60973d4bd6d7466cbe0eac0c"
},
{
"file": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\engineering-evals-ui-7203e02f-ee05-41ab-81dd-40b215086863\\mobile-comparison.png",
"sha256": "0116a8965610d0e2b6a12e6b7ef9f0b46af530dd31e01ea14f2cd50394f585d0"
},
{
"file": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\engineering-evals-ui-7203e02f-ee05-41ab-81dd-40b215086863\\mobile-candidate-evidence.png",
"sha256": "594bbd2c3d8171c3f8e0aa8fadd9f2339fdb4bac8cd6827545433ed81aa21433"
},
{
"file": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\engineering-evals-ui-7203e02f-ee05-41ab-81dd-40b215086863\\desktop-delivered.png",
"sha256": "d67cc3eb5dbaccf3b94dede4f487dd75acf3b071651958362444e9f63107df49"
},
{
"file": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\engineering-evals-ui-7203e02f-ee05-41ab-81dd-40b215086863\\mobile-stopped.png",
"sha256": "6f69b1a0313d9dc7ab33ce21b4a35e83f89a4157be825fe2f753738e20fb6e22"
},
{
"file": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\engineering-evals-ui-7203e02f-ee05-41ab-81dd-40b215086863\\mobile-source-error.png",
"sha256": "926f1c3cafb7690828d7cf1001dfa6d28046895b053897675467b937da9b6aad"
}
],
"exports": [
{
"file": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\engineering-evals-ui-7203e02f-ee05-41ab-81dd-40b215086863\\downloaded-report.json",
"sha256": "fdb56d22b1b5f9778fba08c2ceb0ce57f5d61aed2d19b8ef7f37028c90c895a2"
},
{
"file": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\engineering-evals-ui-7203e02f-ee05-41ab-81dd-40b215086863\\report-8a1d20f4-04ef-421e-8939-342e5422abd0.json",
"sha256": "fdb56d22b1b5f9778fba08c2ceb0ce57f5d61aed2d19b8ef7f37028c90c895a2"
},
{
"file": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\engineering-evals-ui-7203e02f-ee05-41ab-81dd-40b215086863\\report-606d02f2-67b9-443c-bf44-4b36fd764a26.json",
"sha256": "ba603f2c9c8fbdb7ad116e4fcedc17762576258f084179c0bee7bb8cf6586c0c"
}
],
"errors": []
}