Keep Windows Companion aligned with current OpenClaw stable and beta releases

Problem

The Windows Companion and OpenClaw core are separate products, but the Companion owns the supported Windows installation path for the core.

The current Companion Setup Engine still defaults to OpenClaw 2026.6.11, while the current official releases are:

• stable: 2026.7.1
• beta: 2026.7.2-beta.1

This leaves Windows users on an older core when they install through the Companion-owned flow. Avoiding a current stable release because it has a core regression is not a sustainable synchronization strategy: compatibility work and core defects should be tracked and fixed in the repository that owns each behavior.

Goal

Keep the Windows Companion's supported install and upgrade path aligned with current OpenClaw releases:

• update the Companion's default stable OpenClaw version;
• validate the Companion against the current stable core;
• identify compatibility changes needed for the active beta before it becomes stable;
• keep Companion/Windows Node fixes in this repository;
• report and fix OpenClaw core regressions in openclaw/openclaw, rather than carrying Companion workarounds.

Proposed Sequence

1. Land the bounded stable-alignment change in #999:
   • invariant compact-number formatting for pt-PT and other decimal-comma locales;
   • update the Setup Engine LKG from 2026.6.11 to stable 2026.7.1;
   • no unrelated session or /new behavior changes.
2. Prove the current-head Companion installer against stable 2026.7.1 through the real Companion-owned setup path.
3. Classify every reproduced failure by product ownership:
   • Companion/Windows Node defects stay here;
   • OpenClaw core defects move to openclaw/openclaw with independent reproduction.
4. Validate the active 2026.7.2-beta.1 line early enough to prepare compatible Companion changes before the next stable release.
5. Keep version pins, tests, setup documentation, and release evidence synchronized.

Acceptance Criteria

• The Companion no longer defaults to the stale 2026.6.11 core.
• The stable version selected by the Setup Engine is covered by exact tests.
• A real Windows Companion install/upgrade proof exists for the selected stable version.
• Beta compatibility findings are recorded before beta promotion, with fixes made in the owning product.
• No custom OpenClaw fork or silent core workaround is bundled into the Companion.
• Known regressions remain explicit and do not get misrepresented as Companion compatibility fixes.

First PR

• #999 updates invariant pt-PT formatting and the Setup Engine LKG to OpenClaw 2026.7.1.
• The PR remains draft while its current-head installer/runtime proof is completed.
• Known OpenClaw 2026.7.1 startup regression: https://github.com/openclaw/openclaw/issues/108435

Current Status

The first PR has now exercised the real GitHub Actions setup path:

• the main build/test job passed, including Shared, Tray, Connection, Setup Engine, UI, accessibility, and related suites;
• all three setup E2E shards installed OpenClaw 2026.7.1 and then failed before their scenario logic because the Gateway repeatedly reported an active startup-migration lock for the fresh state directory;
• this confirms the version alignment itself and exposes a current OpenClaw core/runtime compatibility blocker;
• #999 remains draft and should not work around that core failure inside the Companion.

Agent Transcript

Start with PR #999 and the Companion Setup Engine's LKG version pin, then run the real GitHub Actions setup path and its setup E2E shards against stable 2026.7.1. Review the reported startup-migration lock and separate Companion/Windows Node failures from the OpenClaw core regression. Done means stable alignment is tested, Windows install/upgrade evidence exists, beta findings are recorded, and no core workaround is added here.