feat(ui): show actual model route and inspect live stream evidence

This commit is contained in:
Agent Workbench Dev
2026-10-31 22:42:47 +08:00
parent eeb11d6a76
commit 31ed219477
22 changed files with 276 additions and 38 deletions
+8
View File
@@ -15,3 +15,11 @@ Responses的typed Items、原call_id配对、原用量与同线路reasoning续
相关恢复检查42/42通过;新增边界29/29、流式/手机投影/上下文38/38通过(这些范围有重叠,不相加)。原日志见[恢复专项](model-protocols-current-specialists.txt)、[边界专项](model-protocols-boundary-specialists.txt)、[投影与上下文](model-protocols-projection-specialists.txt)。新增验证包括原停用对象导致probe引用失效的修复、检测所属attempt精确绑定、原始畸形UTF8捕获、CR事件终结、超时关闭HTTP、自有CCSwitch SQLite/TOML Responses导入,全部使用自有loopback数据。原失败夹具与首次UI定位失败保留;后者为select隐式名称包含option文字,已增加明确aria-label,重新实际验收中。
E界面已构建,正在操作验收;F独立发行与正式部署未执行。协议实现不能代表视觉/音频已接入或真实模型自主工程能力已经通过。具体协议和来源见[协议边界](MODEL-PROTOCOLS.md)。
## E:真实界面与回看修复
模型中心新增协议/能力/范围、实际线路与拒绝原因、完整/未完整流字节与原SHA下载;桌面与伴随端显示同一原Run的有界片段。实际Edge桌面1440px、窄屏390px以及真实TLS网关配对的手机视口5组流程通过,页面错误和失败请求均0,响应下载与工件SHA一致。实际iPhone本轮未联调,不能把窄屏或TLS中继称作实机。入口见[本轮界面结果](model-protocols-e-ui-validation.json)。
目视截图发现并修复:片段同批到达后上游停顿,节流需要定时补发最新状态;选池后主窗口可能仍显示离线演示,旧提交校验还错误要求未选中的原连接模型名。现在用公开选路身份同步主窗口,刷新仍保留实际池名,原连接配置不被改写;新UI流程明确等待选路并刷新后发送。下载直接读取工件字节,畸形UTF8也不再被文本解码改写。37项流式/请求计划/升级、20项旧模型中心/API兼容和26项原字节/连接专项通过(重叠范围不累计)。
原UI定位失败、外部API增加连接后未刷新界面的夹具失败、真实节流缺陷、错误mobile选择器和原首次完整UI运行都保留。首次历史UI全量运行的Purge来源传成打包目录,修正为原source/upstream;原日志与阶段结果单独保存,重跑使用新顶层日志。完整回归早于最后修复启动,属于开发中诊断;最终结论以冻结后的独立源码整轮为准。
+6
View File
@@ -9,3 +9,9 @@
统一消息不能取代真正发送的请求。Responses和流式池各封存实际变体,并另记当前物理请求的SHA/字节;保留原工具call_id、reasoning与原用量。流式进展不进入工具执行,直到原终结与参数通过核对;断流/超时保留unknown不自动重放。原始Buffer捕获保持畸形字节,发生已知凭据脱敏时单列工件SHA。
回顾发现probe引用原可变连接时,toggle会同时更新比较两端,导致旧检测结论被错误接纳;改为连接快照并验证修订。原attempt通过progress身份绑定,避免并发任务改变账本末项。29项新增边界及38项关联投影/上下文检查通过。预算仍按所有成员容量保守取最小值,可能降低混合池可用容量;没有损失请求/审批边界,后续可按真实任务需求优化。E实际验收和F独立发行仍按计划执行。
## E:观察后再判定可用
单看控件存在会遗漏主任务问题。截图中主窗口还显示离线演示,导致发现选池/关闭窗口时的外部状态同步与旧legacy模型校验缺陷;现在选择和刷新都保留实际池名,再由同一后端选路执行。流式首批事件合并时曾只显示0字符,增加定时补发最终片段,停止时清除计时器,专门检查上游不再发送任何事件的情形。
桌面/390px和真实TLS配对观察窗口5组完整流程通过并目视回看。手机原生Safari、信任/PWA已有1.16实机记录;1.17新增流式尚无新实机结论。工程回归继续使用固定原测试和真实Node/React/Python执行,增加同一原工程算法通过Responses SSE与受控Harness的独立门槛,不能拿文本fixture结束当工程完成。
+35
View File
@@ -0,0 +1,35 @@
✔ model registry keeps original credentials/endpoint intact, rejects stale edits, and forgets only volatile keys on restart (2370.3633ms)
✔ config import previews exact revisions, keeps secrets volatile, syncs only when enabled and preserves manual edits on conflicts (2117.705ms)
✔ real HTTP pool fails over 429 to another model and never repeats completed tools; attempts seal actual request bytes and Run identity (6897.5971ms)
✔ pool rotation and weights select members on new Runs; disabled and missing credentials are excluded without requests (9250.5979ms)
✔ 401 blocks the credential revision; updating the key restores availability without leaking error echoes (3217.6818ms)
✔ queued request cancellation and in-flight disable release only owned leases and do not dispatch another HTTP request (2598.7156ms)
✔ compiled model-center HTTP API persists public profiles, blocks forged origins, and local operations never invoke a model (4360.1888ms)
✔ model registry disk failures restore routing, pool profiles and credential references without partial updates (1893.4263ms)
✔ model selection and saved configuration commit together and failures preserve the original mode and endpoint (2166.135ms)
✔ configuration imports roll back the entire batch and retain one usable preview after validation or commit failure (1609.896ms)
✔ HTTP download preserves original malformed SSE bytes and artifact SHA instead of decoding invalid UTF8 (3952.9458ms)
✔ real Chat SSE carries UTF8 and split tool JSON; Runtime executes only after DONE, stores partial progress and original usage (7105.3473ms)
✔ Responses typed SSE validates original item identity, terminal data, raw usage and tool result pair (6291.606ms)
✔ incomplete network stream retains unknown attempt and never automatically switches pool or dispatches fragment tools (4041.245ms)
✔ stopping an active HTTP stream closes only original request and keeps partial text without executing tools (4006.955ms)
✔ explicit streaming probe uses actual SSE despite unknown declaration and binds its own original attempt (2517.4313ms)
✔ coalesced initial SSE events publish the latest partial state even when the upstream stays idle (4539.5055ms)
✔ stream timeout closes the original HTTP response, preserves redacted capture and never retries (2557.65ms)
✔ SSE accepts CR-only final event boundaries and retains exact invalid UTF8 bytes without dispatch (0.9326ms)
▶ stream parser rejects truncation, finish length, malformed JSON, duplicate sequence, changed final item and oversized data
✔ reject 0 (0.1952ms)
✔ reject 1 (0.234ms)
✔ reject 2 (0.2449ms)
✔ reject 3 (5.2352ms)
✔ reject 4 (0.1967ms)
✔ reject 5 (0.1693ms)
✔ stream parser rejects truncation, finish length, malformed JSON, duplicate sequence, changed final item and oversized data (6.8177ms)
ℹ tests 26
ℹ suites 0
ℹ pass 26
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 36601.3161
+41
View File
@@ -0,0 +1,41 @@
{
"success": true,
"version": "1.17.0",
"appRoot": "D:\\Ai-project\\REA逆向\\Agent-Workbench",
"nodePath": "D:\\Ai-project\\REA逆向\\tools\\node-v24.22.1-win-x64\\node.exe",
"containedRuntime": false,
"nodeSha256": "d702daa815eabbff763c7d3808e235cf031cf2ca70211b56847bc72f51e240e5",
"passed": [
"actual form saves Responses and declared stream, refresh preserves both and narrow editor remains operable",
"separate text, nonce tool and actual SSE probes bind only their requested capability and own attempts",
"actual pooled task excludes text-only before dispatch, executes original tool once and downloads SHA-matched original SSE",
"desktop and actual paired gateway show same bounded stream; mobile stops original HTTP with zero fragment tool dispatch or replay",
"truncated output keeps explicit unknown ledger, partial evidence and no automatic failover or incomplete tool execution"
],
"screenshots": [
"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\model-protocols-ui-5c599342-cbbd-4019-8865-ee1d7b4570e4\\desktop-editor.png",
"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\model-protocols-ui-5c599342-cbbd-4019-8865-ee1d7b4570e4\\narrow-editor.png",
"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\model-protocols-ui-5c599342-cbbd-4019-8865-ee1d7b4570e4\\desktop-capabilities.png",
"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\model-protocols-ui-5c599342-cbbd-4019-8865-ee1d7b4570e4\\narrow-capabilities.png",
"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\model-protocols-ui-5c599342-cbbd-4019-8865-ee1d7b4570e4\\desktop-ledger.png",
"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\model-protocols-ui-5c599342-cbbd-4019-8865-ee1d7b4570e4\\desktop-streaming.png",
"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\model-protocols-ui-5c599342-cbbd-4019-8865-ee1d7b4570e4\\narrow-streaming.png",
"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\model-protocols-ui-5c599342-cbbd-4019-8865-ee1d7b4570e4\\mobile-streaming.png",
"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\model-protocols-ui-5c599342-cbbd-4019-8865-ee1d7b4570e4\\mobile-stopped.png",
"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\model-protocols-ui-5c599342-cbbd-4019-8865-ee1d7b4570e4\\narrow-stopped.png",
"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\model-protocols-ui-5c599342-cbbd-4019-8865-ee1d7b4570e4\\desktop-incomplete.png",
"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\model-protocols-ui-5c599342-cbbd-4019-8865-ee1d7b4570e4\\narrow-incomplete.png"
],
"pageErrors": [],
"requestFailures": [],
"fixtureModelRequests": 7,
"realModelRequests": 0,
"osTrustChanged": false,
"nativeIPhone": "not exercised; paired Edge viewport uses CA-verifying owned relay",
"taskId": "238c7b5e-3a50-41bb-a88b-83612d53de00",
"runIds": [
"d9ba991e-8cad-4ee5-bdf1-622f9d71347b",
"67e4674f-57eb-4d97-bd70-2c4ef62eb2ae",
"888065cf-7745-4295-9ae0-cd743c89d682"
]
}
@@ -0,0 +1,46 @@
✔ real Chat SSE carries UTF8 and split tool JSON; Runtime executes only after DONE, stores partial progress and original usage (5983.0173ms)
✔ Responses typed SSE validates original item identity, terminal data, raw usage and tool result pair (6258.0056ms)
✔ incomplete network stream retains unknown attempt and never automatically switches pool or dispatches fragment tools (4047.5165ms)
✔ stopping an active HTTP stream closes only original request and keeps partial text without executing tools (4001.9193ms)
✔ explicit streaming probe uses actual SSE despite unknown declaration and binds its own original attempt (1869.4361ms)
✔ coalesced initial SSE events publish the latest partial state even when the upstream stays idle (4337.4586ms)
✔ stream timeout closes the original HTTP response, preserves redacted capture and never retries (2078.133ms)
✔ SSE accepts CR-only final event boundaries and retains exact invalid UTF8 bytes without dispatch (1.0764ms)
▶ stream parser rejects truncation, finish length, malformed JSON, duplicate sequence, changed final item and oversized data
✔ reject 0 (0.204ms)
✔ reject 1 (0.137ms)
✔ reject 2 (0.192ms)
✔ reject 3 (5.0447ms)
✔ reject 4 (0.7307ms)
✔ reject 5 (0.202ms)
✔ stream parser rejects truncation, finish length, malformed JSON, duplicate sequence, changed final item and oversized data (7.0625ms)
✔ context APIs retain exact request identity, paginate archived groups and reject foreign sessions/runs (6192.0329ms)
✔ delivery reports verify request body, sealed plan and original archived bytes, and show corruption as needing verification (5243.1234ms)
✔ direct code evidence remains selected while recent historical-index browsing is archived within the complete request budget (4520.4396ms)
✔ one long user turn becomes bounded complete groups with exact UTF-8 history and recent JSON arguments (10567.6007ms)
✔ tool definitions, exact system bytes, output reserve and remaining Run/Task budget are accounted before requests (2088.1285ms)
✔ oversized required user instructions pause intact and can be explicitly resumed after increasing the window (4121.4042ms)
✔ all task instructions, constraints and acceptance remain in the runtime system instead of an eight-item or byte prefix (3632.5053ms)
✔ a newest manageable multi-tool group stays exact even beyond the working target; oversized earlier groups are archived (2187.4153ms)
✔ invalid orphan result is rejected before archive or network preparation (1310.8086ms)
✔ invalid missing result is rejected before archive or network preparation (1405.7107ms)
✔ invalid wrong result is rejected before archive or network preparation (1532.416ms)
✔ invalid duplicate call is rejected before archive or network preparation (2103.016ms)
✔ a group larger than the artifact limit uses verified Unicode-safe chunks without replacing tool JSON (4459.0071ms)
✔ changed or missing archived evidence stops reuse and is never silently recreated (2227.314ms)
✔ transaction failure prevents publishing a plan and any corresponding model request (2090.0617ms)
✔ the Provider rejects body mutation after a sealed plan before calling HTTP (1741.8434ms)
✔ HTTP failure and explicit same-Task resume reuse exact completed results and request hashes without repeating a command (13863.0813ms)
✔ a real process exit after plan persistence marks interruption and never issues an automatic model request (2513.3048ms)
✔ hot upgrade adopts a retained virtual data root without reopening SQLite or changing rows, Provider or routing (2472.0281ms)
✔ failed adoption restores the original virtual data-root name and original open database (1811.855ms)
✔ hot upgrade refreshes retained model and importer methods while preserving credential and routing objects (2029.5545ms)
✔ hot upgrade refuses active probes and restores retained model methods if installation fails (1963.2231ms)
ℹ tests 37
ℹ suites 0
ℹ pass 37
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 72015.918
@@ -0,0 +1,28 @@
✔ the bundled local service enforces host, origin, token and JSON object contracts (2974.454ms)
✔ an in-flight connection probe prevents configuration changes and another probe until its actual result returns (4037.4954ms)
✔ API review applies verified content, detects conflicts, exports sessions and keeps keys in memory (7025.1254ms)
✔ a real bundle restart preserves proposals, interrupts old approvals and requests fresh consent on resume (20310.8193ms)
✔ runtime diagnostics and model-response validation use actual records and preserve the saved connection (3695.7218ms)
✔ persistent event replay is read-only and keeps run/artifact identity across reconnects (13297.8198ms)
✔ model registry keeps original credentials/endpoint intact, rejects stale edits, and forgets only volatile keys on restart (1701.7823ms)
✔ config import previews exact revisions, keeps secrets volatile, syncs only when enabled and preserves manual edits on conflicts (1223.0659ms)
✔ real HTTP pool fails over 429 to another model and never repeats completed tools; attempts seal actual request bytes and Run identity (6081.7127ms)
✔ pool rotation and weights select members on new Runs; disabled and missing credentials are excluded without requests (8632.154ms)
✔ 401 blocks the credential revision; updating the key restores availability without leaking error echoes (4455.0191ms)
✔ queued request cancellation and in-flight disable release only owned leases and do not dispatch another HTTP request (1923.0023ms)
✔ compiled model-center HTTP API persists public profiles, blocks forged origins, and local operations never invoke a model (4051.2999ms)
✔ model registry disk failures restore routing, pool profiles and credential references without partial updates (1738.0266ms)
✔ model selection and saved configuration commit together and failures preserve the original mode and endpoint (1983.278ms)
✔ configuration imports roll back the entire batch and retain one usable preview after validation or commit failure (2341.2948ms)
✔ hot upgrade adopts a retained virtual data root without reopening SQLite or changing rows, Provider or routing (2644.1625ms)
✔ failed adoption restores the original virtual data-root name and original open database (1229.9266ms)
✔ hot upgrade refreshes retained model and importer methods while preserving credential and routing objects (1692.9317ms)
✔ hot upgrade refuses active probes and restores retained model methods if installation fails (1448.422ms)
ℹ tests 20
ℹ suites 0
ℹ pass 20
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 51433.0448