Anna Workbench RC2 · Source Developer Preview

RC2 fixes the release-validation failures exposed in RC1 while preserving the shared Home/Create, Cowork, and Crew conversation flow.

RC2 修复 RC1 暴露的测试超时与 CI 证据依赖问题，保留持续会话、能力调用、定向停止及既有业务入口。本次为 macOS arm64 源码 Developer Preview，没有新的安装包、签名或公证。

Source / 最终提交: e2e603cbc70de5fddaeb4036e001cbaba4e4da0c

Changes / 本轮修复

• Two workdir registration/admission tests no longer initialize OMP for operations that finish before a Run starts. Timing probes reproduced the original 5-second timeout and isolated repeated runtime-manifest scanning as the dominant setup cost. Real Python identity, Product Host, persistent Runtime, and original assertions remain; the two tests that execute file reads still use real OMP. Production integrity checks and test timeouts are unchanged.
• CI checks continue after an unrelated failure when their direct dependencies are ready. Python installation and tests remain independent of npm failures. Evidence verification runs after successful generation or when a manifest exists; uploads require actual evidence. Failed generation remains visible, partial evidence can be retained, and cancellation stops downstream work.
• English and Chinese READMEs describe RC2, retain existing capabilities and Hiker authorship/license boundaries, and link the new validation record. The original RC1 failure record is preserved.

两个目录登记/入场拒绝用例保留真实身份与范围校验，移除未执行模型时的 OMP 初始化；CI 补齐实际依赖与取消条件，避免缺文件连锁错误。各项测试时限与原有断言保持不变，未新增 skip。

Validation / 验证

Local validation used macOS arm64, Node 24.12.0 and Python 3.12.13. The final adjustment changed only the CI job budget and its documentation; the tested production and test inputs, runtime packages, and lockfiles remained identical, so the full local suites were not repeated.

Check	Result
Full JavaScript suite	1241 passed, 0 failed, 7 pre-existing skipped, exit 0
Full Python suite	1093 passed, exit 0
Typecheck	Passed
Web and Harness Host builds	Passed
Maintained frontend product smoke	9 passed
Real Product Host/OMP browser UI regression	1 passed, about 41 seconds; screenshot inspected
CI dependency guard	3 passed, including failed prerequisites, missing/partial outputs, and cancellation
Public release boundary scan	Passed
Independent Standards / Spec reviews	0 remaining findings on each axis

Exact-commit CI: run 34763449213 — checks: success; product: success. Every step completed successfully, including UI/Python and WB00/WB01/WB02 generation, verification, and upload. All three evidence artifacts are retained.

The initial RC2 candidate run passed its full JavaScript suite in 26 minutes 45 seconds and passed the product job, but GitHub cancelled the checks job at its 30-minute total limit during UI validation. That failed release gate remains recorded. The checks budget is now 45 minutes to accommodate the full required sequence; individual test limits remain unchanged.

The product CI job includes the macOS arm64 application build and branded packaged-executable smoke. That check does not establish signed distribution or Windows/Linux acceptance.

Limits / 未完成边界

• The UI regression uses real local identity, stores, Gateway, and OMP with a fixed external model transport. It proves local behavior; current live Provider/MCP acceptance remains blocked. Usage stays unavailable where the provider does not report it.
• Full scheduling, steer, ask/answer, crash recovery, directory/content search, additional business capability catalog entries, Memory, Sandbox, and general execution remain pending. Full M1 and production acceptance remain open.
• Hiker read tests do not establish live writes/readback. Existing Hiker permissions and reimbursement/Crew state machines remain authoritative. Hiker retains its own authorship, private-source, and license boundaries.
• This release contains source only. It does not deliver a new installer, signing/notarization, or Windows/Linux acceptance.

真实 Provider/MCP、完整 M1、签名分发、Windows/Linux 与生产验收仍未完成。合成 transport 与本地 UI 测试不代表真实 Hiker 写入/读回已通过。

See the RC2 record and quick start. RC1 and its failed checks run remain unchanged historical records.