diff --git a/platform-validation/native-linux.md b/platform-validation/native-linux.md new file mode 100644 index 00000000000..cd72ba555d9 --- /dev/null +++ b/platform-validation/native-linux.md @@ -0,0 +1,10 @@ +# Native Linux/package validation + +- Branch/head: `brennanb2025/structured-chat-integration-latest` / `be52c0d6dcd9478f0303335de5eda5224039e82b` +- Available evidence: Linux structured lifecycle, journal, prompts, options, images, resume, + close, and unexpected-exit behavior is covered by the passing focused suites and the exact-head + SSH/Linux run. + +Packaged Linux startup, real provider-process smoke, and glibc 2.31-floor validation were not +available in this session. This platform therefore remains **land after fixes/evidence** rather +than merge-ready. diff --git a/platform-validation/paired-remote-orca.md b/platform-validation/paired-remote-orca.md new file mode 100644 index 00000000000..f706caedcfa --- /dev/null +++ b/platform-validation/paired-remote-orca.md @@ -0,0 +1,15 @@ +# Paired remote Orca validation + +- Branch/head: `brennanb2025/structured-chat-integration-latest` / `be52c0d6dcd9478f0303335de5eda5224039e82b` +- Client: local desktop Orca. +- Runtime: paired **Windows high spec**, runtime id `4b5a97ba-c5cf-4c58-9029-6dff541b788b`, app + `1.4.193-adhoc.20260830235404`. + +The client reported runtime state `ready` and capabilities including `agent-session.structured`, +`agent-session.host-authority`, shared-control, and runtime-environment support. The exact branch +was checked out on the paired host and exercised through the paired terminal. Windows structured +RPC/router/runtime tests passed (38 tests), and no provider process was launched by the local +client during validation. + +Live mixed-version publication/reconnect, runtime-unavailable versus transport-connected UI, and +an authenticated provider turn were not performed. Keep this topology **land after fixes/evidence**. diff --git a/platform-validation/ssh-linux.md b/platform-validation/ssh-linux.md new file mode 100644 index 00000000000..1c123315189 --- /dev/null +++ b/platform-validation/ssh-linux.md @@ -0,0 +1,15 @@ +# SSH/Linux structured-chat validation + +- Branch/head: `brennanb2025/structured-chat-integration-latest` / `be52c0d6dcd9478f0303335de5eda5224039e82b` +- Host: OpenClaw SSH execution host `neil-ubuntu`, Linux `7.0.0-28-generic`, x86_64. +- Checkout: `/home/brennan/orca-structured-chat-ssh-exact`. + +## Commands and results + +- Structured adapter/runtime tests: passed; 37 tests. +- SSH verdict, reconnect, and process-liveness suites: passed; 19 tests. +- `command -v claude; claude --version; command -v codex; codex --version`: Claude Code `2.1.233`, Codex CLI `0.151.0`. + +The tests preserve the exact SSH verdict vocabulary `live`, `unverifiable`, and `exited`, and +assert that transport loss is not process death. A live authenticated provider turn and a forced +network disconnect/reconnect against a running structured provider were not performed. diff --git a/platform-validation/windows.md b/platform-validation/windows.md new file mode 100644 index 00000000000..813dcae8c36 --- /dev/null +++ b/platform-validation/windows.md @@ -0,0 +1,23 @@ +# Windows structured-chat validation + +- Branch/head: `brennanb2025/structured-chat-integration-latest` / `be52c0d6dcd9478f0303335de5eda5224039e82b` +- Host: Orca paired **Windows high spec**, `win32`, Node `24.18.0`, x64. +- Checkout: `C:/Users/neil/orca/workspaces/orca/structured-chat-windows-exact`. + +## Commands and results + +- `pnpm test src/main/windows/windows-process-table.test.ts src/main/wsl/wsl-executable-path.test.ts src/main/runtime/orca-runtime.test.ts`: passed; 25 process-table tests and 1,221 runtime tests. +- `pnpm test src/main/claude`: passed; 41 files, 344 passed, 11 skipped. +- Structured RPC/router/runtime suites: passed; 38 tests. +- Provider discovery: Claude Code `2.1.251`; Codex CLI `0.151.0`. + +The process-table capability gate and native process addon were exercised on Windows. Native +Claude structured launch is intentionally refused on `win32`; the caller receives the truthful +unsupported result and retains the terminal fallback. A live authenticated provider turn, +packaging, and path-length run were not performed. + +## WSL limitation + +`wsl.exe -l -q` listed `Ubuntu-24.04` and `Sta4593-Federated`. Starting `Ubuntu-24.04` failed +with `Wsl/Service/CreateInstance/MountDisk/HCS/ERROR_SHARING_VIOLATION` because its VHDX was in +use, so provider execution inside WSL remains unverified. diff --git a/structured-chat-readiness-report.md b/structured-chat-readiness-report.md index 773a206de9d..3cd449201e6 100644 --- a/structured-chat-readiness-report.md +++ b/structured-chat-readiness-report.md @@ -1,7 +1,7 @@ # Structured native-chat readiness report Integration branch: `brennanb2025/structured-chat-integration-latest` -Implementation/test head (code): `62521d69c6c0472d94695b5e92e6739fb9de7589` +Implementation/test head (code): `be52c0d6dcd9478f0303335de5eda5224039e82b` ## Architecture decision @@ -25,16 +25,25 @@ the recoverable legacy terminal path is retained. - **Native macOS/Linux:** provider lifecycle, journal, prompts, options, images, resume, close, and unexpected-exit recovery are covered by passing suites. Packaged provider-process smoke was not run here. -- **Windows:** structured capability is advertised only when the process-table addon proves - creation-time identity; otherwise the renderer stays on terminal chat. Real Windows native and - fallback execution, packaging, and path-length CI remain unverified. +- **Windows:** exact pushed head was checked out on the Windows high-spec paired runtime. The + process-table suite (25 tests), runtime suite (1,221 tests), Claude suite (41 files, 344 passed, + 11 skipped), and structured RPC/router suites (38 tests) passed. Claude native structured launch + remains intentionally unsupported on native Windows and is covered by the truthful capability + fallback; `Claude Code 2.1.251` and `codex-cli 0.151.0` were present. Packaging, path-length, + and a live authenticated provider turn remain unverified. - **WSL:** absolute executable resolution, distro-aware routing, WSL git probes, and location - tests pass. A selected-distro provider attach/teardown run was unavailable. -- **SSH/Linux:** verdict vocabulary remains exactly `live`, `unverifiable`, `exited`; transport - loss never proves death. Real OpenClaw disconnect/reconnect evidence is unavailable. -- **Paired remote Orca:** owning-host target resolution, publication/lifecycle routing, and - cross-version unit coverage pass. A paired-host Electron run and live mixed-version run were - unavailable. + tests pass. The selected-distro probe on the Windows runtime listed `Ubuntu-24.04` and + `Sta4593-Federated`, but attach failed with `ERROR_SHARING_VIOLATION` while mounting the + Ubuntu VHDX; provider-in-WSL execution remains unverified. +- **SSH/Linux:** exact pushed head ran on `neil-ubuntu` (Linux 7.0.0-28, x86_64): structured + adapter/runtime tests passed (37), SSH verdict/reconnect/liveness suites passed (19), and + Claude 2.1.233 plus Codex 0.151.0 were present on the execution host. A live authenticated + provider turn and forced network disconnect/reconnect remain unverified. +- **Paired remote Orca:** the desktop client reached the Windows high-spec paired runtime at + app version `1.4.193-adhoc.20260830235404`; runtime state was `ready` and it advertised + `agent-session.structured`, `agent-session.host-authority`, and shared-control capabilities. + The exact branch was checked out and exercised through the paired terminal; live mixed-version + publication/reconnect and authenticated provider-turn evidence remain unverified. - **Electron:** exact-head Playwright CDP proof passed from the integration worktree. The app identity reported `structured-chat-integration-final` on this branch; Experimental settings visibly exposed the Chat UI and structured-chat controls, the host-aware scope copy rendered,