Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 5 additions & 0 deletions docs/3.0/PROGRESS.md
Original file line number Diff line number Diff line change
@@ -1,5 +1,10 @@
# PROGRESS · Clawket 3.0 进度日志

- 2026-10-16 Native 有界流开发包真实复验与第一轮 UI 初验准备(RN 保持默认)。
- #276 exactfe157111/current-main/CLEAN全部11门禁绿后合2d8a7890并finish;仅证据文档,不改产品。新独立acceptance工作树使用#277 productiond5d2e6a3(与重基6d2fd826源码/测试等价)开发App,不打Archive/IPA或改共享服务。
- 实际UI automation stale导致预检停在Send0/Stop0,临时Studio草稿已恢复,三refs/held保留;随后真saved connection、header/原draft/导航route精确核对。新真实Send1经被动reader绑定receipt/见非空live,Native单assistant15630字节、streaming=false、无failure/Stop,长流完整结束。元素Stop调用未拿到中止证据;另一次新input单Send/单coordinate tap、live截图有Stop,最终亦final非aborted。未证明handler或坐标等于live center,原生取消仍未通过,均未重发。
- 真实系统Paste恢复原16字draft/0refs、Send可用;独立定向history核首条自有输入/samephysical/noactive。不是Relay/弱网/稳定性或Stop通过。初验聚焦聊天/编辑/附件/预览/导航,真实Hermes文本另验;OpenClaw认证、Stop、网络恢复/VoiceOver/完整N5继续待验,默认客户端不切换。

- 2026-10-16 Native Runtime 消费驱动事件缓冲(独立修复;RN 保持默认)。
- 保留公开 AsyncThrowingStream,内部单一64项/16MiB有界队列随真实next出队;仅同内部代际且UTF8精确路由/segment/lineage、无snapshot metadata的相邻纯delta合并,其他事件保持barrier。原始事件weight全部计费,容量溢出仍同步退休、不丢已准入正文、不提高限额、不自动重放;取消/finish单次通知并唤醒reader。
- 两个独立SPM文件覆盖慢consumer千片Unicode/空白、真实Hermes/Pi mapper→Coordinator→公开stream、tool/work/snapshot/final顺序、Stop不等UI消费、严格身份/代际隔离、65个非合并项、字节限/超限、取消/finish/owner释放与pending写退休无重放。首编strict Swift6/jobs2 39.21秒,新buffer9方法/10输入与真实runtime5方法首轮全绿;既有SessionLifecycle5方法/11输入、RuntimeOperations6方法/12输入逐文件全绿,16指令对/7docs与diff通过。strict Swift6/iOS16.4双arch实际App开发编译/codesign验证也已通过,私有dev副本交主线程,heavy已释放;不属发行包或设备验收。独立只读review与完整CI并行。
Expand Down
10 changes: 10 additions & 0 deletions docs/native-ios/BACKEND-ACCEPTANCE.md
Original file line number Diff line number Diff line change
Expand Up @@ -42,6 +42,16 @@ The synchronous debugger now actually stopped in `showFailure`, with its caller

The same app's later standard-text-size Studio review passed actual photo preview/open/Done with a zero-point change in the visible code button's reading position, Copy feedback plus the exact 45 code bytes including the trailing newline, explicit Latest, and Work/Done/Menu/Connection settings/Done. The background WebDriver pasteboard API returned empty; the owned simulator's pasteboard read independently matched the known fixture, so the empty automation read is not recorded as a Copy defect. The real QA draft and zero resources, Studio's original draft and three ordered resources, and its one held item were all preserved, with no queued or sending items. Studio is a synthetic UI environment; these checks do not establish real inference, backend Stop, VoiceOver or recovery stability.

## Runtime-buffer development app recheck

The `d5d2e6a3` development app (production sources equal to rebased `6d2fd826`) was installed over the same QA sandbox. A zero-send preflight stopped after stale UI automation; the temporary Studio draft was restored before selecting the actual saved Hermes connection. The real header, original draft and exact stored QA route were then verified. No held Studio input or attachment was sent.

One fresh real input received a run receipt and independently observed live text. Native retained one 15,630-byte assistant reply, reached `streaming=false`, removed Stop and showed no failure notice. The element-based Stop automation exited without an authoritative abort receipt; the independent terminal was `final`. A separate fresh input used a pre-Send control coordinate and a live screenshot containing Stop, then one coordinate-tap command. That run also ended `final`, not `aborted`. Neither trial is successful cancellation; the latter does not establish handler entry or the coordinate's equality with the live button center. Further Stop diagnosis must capture that boundary rather than replay either input or infer backend cancellation from the UI.

Afterward, there was no failure notice or Stop control. Actual system Paste restored the original 16-character draft with zero resources and Send enabled. An independent read-only history check verified the original owned input, same physical session and no active run. This establishes a completed real long reply and the later idle window; it is not full-network stability, an original-failure typed sample or successful native cancellation.

The first owner review may focus on Studio's chat typography, code Copy, scroll/Latest, multiline editing, ordered three-image tray, preview/return and Work/Menu navigation. Studio is synthetic; use the separate Hermes QA conversation for real text only. OpenClaw authentication, native Stop, Relay/network recovery, VoiceOver and full parity remain pending. React Native remains the default.

## Later authentication diagnostics

On the `0f648ece` development app, a private loopback application-frame observer forwarded original frames without adding credentials or issuing RPCs. It classified OpenClaw's response as `INVALID_REQUEST` / `AUTH_TOKEN_MISMATCH`, and an in-memory comparison found that the submitted `connect.auth.token` differed from the current local operator configuration. A fresh form with explicit secure-field focus, slower WebDriver input and transition waits produced the same classification. The secure field does not expose the literal value through accessibility, so its masked value does not establish which input reached the binding. Form entry, automation and credential selection still require diagnosis; no token or service configuration was changed.
Expand Down
Loading