Crabbox 0.11.0 adds Google Cloud provider and repo-local job workflows

• steipete's Crabbox 0.11.0 post says the release adds four concrete pieces: a Google Cloud provider, repo-local job workflows, AWS Windows WSL2 hydration, and a Blacksmith sync-stall guard.
• In steipete's bug-repro workflow, Crabbox is the disposable environment around Codex: recreate the bug, verify it, apply the fix, then verify again, with up to 10 parallel sessions.
• steipete's OpenClaw QA post shows Crabbox being used to generate before-and-after proof artifacts, including screen recordings for a real Telegram login flow.
• steipete's Windows terminal demo adds one more useful signal about the platform work: recent terminal handling was good enough for an end-to-end fix that renders animated GIFs inside a Windows terminal.

You can jump straight to the Crabbox v0.11.0 release page, inspect the OpenClaw proof-generation example, and browse the public Crabbox site. The interesting part is how quickly the release bullets line up with live usage: ephemeral bug repro on Friday, proof-generating QA on Saturday, then the 0.11.0 feature drop on Sunday.

Crabbox 0.11.0 is a small release note with a pretty clear theme: more places to run jobs, and more guardrails around those runs. In steipete's release thread, the four shipped items break down like this:

• Google Cloud provider
• Repo-local job workflows
• AWS Windows WSL2 hydration
• Blacksmith sync-stall guard

The first two expand where Crabbox jobs can execute and how closely they can live with a repo. The second two are more operational, one aimed at Windows environment setup, the other at avoiding a specific sync stall during Blacksmith-backed work. The linked GitHub release is the canonical reference for the version.

The strongest evidence for what Crabbox is for comes from steipete's workflow post, not the changelog. He describes a loop where Codex recreates the exact bug state inside an ephemeral Crabbox, verifies the issue, applies a fix, and verifies the fix again.

The attached screenshot in that post is unusually concrete. It lists the changed files, the regression tests, sanity checks like git diff --check, and two distinct Crabbox runs, one for repro and one for fix verification. The other operational detail is parallelism: Steinberger says he runs 10 sessions at once, which frames Crabbox less like a one-off sandbox and more like disposable CI for agent-driven debugging.

The OpenClaw example pushes Crabbox a step past repro and into audit trail generation. According to steipete's post, Codex or a GitHub workflow generates before-and-after proofs, while Crabbox handles the screen recording.

The screenshot in that proof post shows a side-by-side Telegram comparison, with proof notes that call out the exact formatting delta between main and the PR head. Steinberger also says the flow includes real Telegram login automation, and the linked GitHub comment is where the full proof artifact lives.

The last useful breadcrumb is platform maturity. In steipete's demo, he says Crabbox's Windows terminal handling is now good enough that Codex could end-to-end fix gifgrep so animated GIFs render directly in the terminal.

That sits neatly beside the 0.11.0 note about AWS Windows WSL2 hydration from the release post. One is a release bullet, the other is a visible proof that Windows support is getting exercised in real agent runs, not just listed in a changelog.