Structured code review through the shared OpenClaw agent-skills installation.
日本語の概要は準備中です。原文の説明を表示しています。
Uninstall or hard-clean OpenClaw Companion, Windows node, native Gateway/MXC, WSL Gateway, and managed llama.cpp/Local AI state for a clean retest. Choose the existing dev CLI uninstall or the reviewed hard-clean procedure, inventory exact targets, and obtain destructive confirmation before acting.
インストール方法を見るインストールする前に、エージェントに与えられる指示の中身を確認できます。
Creating or updating cleanup tooling is not permission to uninstall anything. Inventory first, show the exact proposed targets and preserved resources, then ask for confirmation before stopping processes or changing installed state.
| User intent | Procedure |
|---|---|
| Exercise the dev build's WSL uninstall engine | Dev-build engine uninstall below. |
| Clean reinstall, hard wipe, remove native Gateway/MXC or all managed llama.cpp state | Read HARD-CLEAN.md before acting. The dev engine path alone is not a native hard clean. |
| Demo/test device without Copilot | Copy scripts\clean-uninstall.ps1 to the device. Run with Windows PowerShell 5.1; default is preview, -ConfirmDestructive applies. See standalone usage. |
| Ordinary installed-app removal | Settings > Apps > Installed apps. See docs\SETUP.md, Uninstalling. Retained state is intentional. |
The standalone script's -All flag enables dev state, receipt-backed cached models,
and owned release/dev WSL distro removal. It does not imply -ConfirmDestructive
or discover custom profile paths. Review and confirm the expanded scope first.
Use -All -ExcludeCachedModels to preserve external shared cached models while
selecting the other built-in scopes. The exclusion overrides -RemoveCachedModels
too; models inside deleted profiles or WSL filesystems are still removed.
For explicitly identified leftover Windows isolated-profile registrations, use
-RemoveIsolatedProfilePath with exact paths. Preview first; confirmed removal
requires elevation as the same user and accepts only unloaded, non-special
S-1-5-110 profiles. -All never selects these automatically. See HARD-CLEAN.md
for the distinction between profile cleanup and native session teardown.
Existing helpers are narrower than a hard clean:
scripts\dev-reset-rebuild-launch.ps1 resets production tray state and can wipe
the production WSL distro. It mutates by default; use -WhatIf to preview.scripts\reset-openclaw-wsl-validation-state.ps1 is a production-WSL validation
reset with backups and a -ConfirmDestructiveClean gate. It is not a native
Gateway/MXC package teardown.scripts\_uninstall-helpers.ps1 supplies utilities, not a complete uninstaller.Do not chain these helpers together as an "all" wipe. Their identities, backup behavior, process selection, and WSL scope differ.
OpenClaw.Tray.WinUI.exe embeds the SetupEngine and accepts --uninstall directly, so
no installer is required to exercise this path; a local dotnet build output works.
This walks a real (not just installed-via-Inno-Setup) uninstall of dev/test state:
WSL gateway distro, openclaw CLI inside WSL, gateway systemd service, Tailscale
state, and the tray's AppData files.
This is destructive. Confirm with the user before the final --confirm-destructive
run, and call out anything it will touch that isn't disposable dev state (e.g. a WSL
distro or Tailscale session actually in use).
Build with the dev identity, for the matching architecture. .\build.ps1
alone defaults to release identity even in a Debug configuration, and a
release-identity exe would unregister the real OpenClawGateway distro and mutate
%APPDATA%\OpenClawTray instead of the -Dev copies below (the output path is
identical either way, so nothing later in this procedure would catch that
mistake). Always build with -DevBuild, and verify the marker before proceeding:
.\build.ps1 -DevBuild # skip only if you already built with -DevBuild
$arm64 = [System.Runtime.InteropServices.RuntimeInformation]::OSArchitecture -eq
[System.Runtime.InteropServices.Architecture]::Arm64
$arch = if ($arm64) { 'win-arm64' } else { 'win-x64' }
$outDir = "src\OpenClaw.Tray.WinUI\bin\Debug\net10.0-windows10.0.22621.0\$arch"
$exe = "$outDir\OpenClaw.Tray.WinUI.exe"
$identity = (Get-Content "$outDir\app-identity.txt" -Raw).Trim()
if ($identity -ne 'dev') { throw "Build output identity is '$identity', not 'dev' - rebuild with -DevBuild." }
$pathOverrides = 'OPENCLAW_TRAY_DATA_DIR',
'OPENCLAW_TRAY_APPDATA_DIR',
'OPENCLAW_TRAY_LOCALAPPDATA_DIR',
'OPENCLAW_TRAY_LOCAL_DATA_DIR',
'OPENCLAW_STATE_DIR'
$activeOverrides = $pathOverrides | Where-Object {
-not [string]::IsNullOrWhiteSpace([Environment]::GetEnvironmentVariable($_))
}
if ($activeOverrides) {
throw "Refusing destructive uninstall while path overrides are set: $($activeOverrides -join ', ')"
}
Stop only the dev instance, by PID. A background OpenClaw.Tray.WinUI.exe
(e.g. the tray icon) holds its own log/journal files open, which makes later steps
fail to delete the AppData Logs directory. A release-identity tray can share the
same process name, so match on the resolved $exe path and confirm before
stopping; this repo kills processes by PID only, never by name
(see scripts/dev-reset-rebuild-launch.ps1):
$exePath = (Resolve-Path $exe).Path
$devProcs = Get-Process OpenClaw.Tray.WinUI -ErrorAction SilentlyContinue |
Where-Object { $_.Path -eq $exePath }
$devProcs | Format-Table Id, Path -AutoSize # confirm these are the dev instance before stopping
$devProcs | ForEach-Object { Stop-Process -Id $_.Id -Force }
Dry-run first, capturing output and exit code (a bare shell invocation can swallow
the exit code for a WinUI apphost; use Start-Process -Wait -PassThru):
$p = Start-Process -FilePath $exe -ArgumentList '--uninstall','--dry-run' -PassThru -Wait `
-RedirectStandardOutput out.log -RedirectStandardError err.log
$p.ExitCode
Get-Content out.log
Review the "Would rollback: ..." lines for each of the ~37 steps. Note the
--data-dir / --distro-name printed at the top; dev-branch builds use
OpenClawTray-Dev / OpenClawGateway-Dev, separate from a real install's
OpenClawTray / OpenClawGateway. Confirm that the printed data directories
are exactly %APPDATA%\OpenClawTray-Dev and
%LOCALAPPDATA%\OpenClawTray-Dev. Stop if they point anywhere else.
Check for real state the dry-run doesn't surface, e.g. a live WSL distro:
wsl -l -v
If the distro named in step 3's output is present, flag to the user that it will be
unregistered (wsl --unregister), permanently deleting its filesystem.
The dry-run only logs rollback step IDs, not the concrete files each step would
touch. If it matters to the user's situation, inspect %APPDATA%\OpenClawTray-Dev\
directly before consenting (gateway registry, windows-node-context.json, Local AI
config, Tailscale state).
Confirm with the user, then run the real uninstall the same way:
$p = Start-Process -FilePath $exe -ArgumentList '--uninstall','--confirm-destructive' -PassThru -Wait `
-RedirectStandardOutput out2.log -RedirectStandardError err2.log
$p.ExitCode
Get-Content out2.log
Verify the specific artifacts each rollback step is supposed to clean up: not the
AppData roots themselves. TrayArtifactCleanup (registry/files/settings only, no WSL
calls) and the WSL/gateway rollback steps between them reset or remove selected files
and records in place; neither deletes %APPDATA%\OpenClawTray[-Dev] or
%LOCALAPPDATA%\OpenClawTray[-Dev], so asserting those roots are gone will report a
successful run as failed:
wsl -l -v # target distro should no longer be listed
# The current run's own log/journal remain under Logs\Setup (they are open during
# cleanup); success = that directory holds only uninstall-engine-<this-run> files
Get-ChildItem "$env:APPDATA\OpenClawTray-Dev\Logs\Setup" # only this run's uninstall-engine-* files expected
Test-Path "$env:LOCALAPPDATA\OpenClawTray-Dev\Logs" # should be False
Test-Path "$env:LOCALAPPDATA\OpenClawTray-Dev\run.marker" # should be False
Test-Path "$env:APPDATA\OpenClawTray-Dev\exec-approvals.json" # should be False
Test-Path "$env:LOCALAPPDATA\OpenClawTray-Dev\windows-node-context.json" # should be False, if it existed
# settings.json: GatewayUrl removed; EnableNodeMode/AutoStart reset to false unless
# other gateway records remain
# Get-ItemProperty 'HKCU:\SOFTWARE\Microsoft\Windows\CurrentVersion\Run' -Name OpenClawTray-Dev -ErrorAction SilentlyContinue
# should error/be absent (autostart registry key removed)
The current run's own active log and journal under %APPDATA%\OpenClawTray-Dev\Logs\Setup
are expected to remain: the Setup Engine still holds them open when
TrayArtifactCleanup runs, so the Failed to delete AppData Logs directory warning
is normal. A Failed to delete... warning naming any other file, or leftover files
from earlier runs, means a process is still holding them open (see step 2).
%APPDATA%\OpenClawTray[-Dev]\Logs\Setup\uninstall-engine-<timestamp>.jsonl.docs/SETUP.md#uninstalling
and the README's Uninstall section. The procedure above exercises the underlying
CLI path directly. For a native or full clean-retest workflow, use
HARD-CLEAN.md.まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Structured code review through the shared OpenClaw agent-skills installation.
日本語の概要は準備中です。原文の説明を表示しています。
Use Crabbox from macOS, Linux, or native Windows controllers to run OpenClaw Windows node builds, tests, and targeted proof on remote native Windows or WSL2 hosts, including Azure or brokered AWS leases and static SSH hosts. Use when remote Windows validation is needed or the user asks for Crabbox validation. Always report the actual provider, lease id, run URL, command, and result.
日本語の概要は準備中です。原文の説明を表示しています。
Run complete OpenClaw Windows Node issue and PR triage, including landing queues, proof pools, ownership audits, and release planning.
日本語の概要は準備中です。原文の説明を表示しています。
Adversarial dual-model code review that runs two AI models in parallel, cross-references findings, and produces a consensus severity table with fix confidence ratings. Use when asked for a thorough code review, adversarial review, dual-model review, confidence-rated review, or 'Hanselman review'.
日本語の概要は準備中です。原文の説明を表示しています。
Build the native (MSIX, isolated-session) OpenClaw Gateway from the latest openclaw/openclaw sources (or any ref, local checkout, or prebuilt openclaw.tgz), register it side by side with the Microsoft Store Gateway, and run Companion against it with OPENCLAW_NATIVE_GATEWAY_DEV_PATCH. Use when the user asks to test Companion with Gateway main, an unreleased Gateway change, or a source-built native Gateway, or to remove such a build.
日本語の概要は準備中です。原文の説明を表示しています。
Plan and collect OpenClaw Windows validation/proof: tests, rubber-duck review, UI evidence, MCP output, and gateway runtime proof.
日本語の概要は準備中です。原文の説明を表示しています。