[jarrad]

Summary

• Keep the existing brev start duplicate remediation when workspace creation returns HTTP 400 and Brev code WORKSPACE_ALREADY_EXISTS.
• A generic 400, a different code, or the same code on HTTP 500 keeps ordinary error handling. Message text no longer triggers duplicate guidance.
• Stop the creation spinner when the request fails.

The old pkg/cmd/create implementation is explicitly deprecated and is not registered as brev create; the registered command is pkg/cmd/gpucreate, whose retry handling belongs to BREV-13207.

Proof

• New start regression failed on the baseline: coded duplicate lacked guidance, while unrelated duplicate-like text received it. It passes with this change.
• go test ./pkg/cmd/start -count=1 passed.
• go test ./pkg/store -run '^(TestHTTPResponseError|TestIsHTTPErrorCode)' -count=1 passed.
• go vet ./pkg/store ./pkg/cmd/start passed.
• Full go test ./pkg/store has two existing WSL-specific tests that fail on macOS (TestBasicStore_GetWindowsDir, TestFileStore_GetWSLHostUserSSHConfigPath).

The stacked PR targets a branch, while the repo's automatic PR workflows filter for main. I dispatched the non-deployment validation workflows on this exact head (9ccc9928c8f4e61eab698e27f8091681f8b2f263): test, lint, build, vet, fmt, and CLI output compatibility. All passed. GitHub does not show these manually dispatched runs in the PR check rollup.

Stack and rollout

Base: CLI parser foundation #515 (fix/brev-13209-cli-error-codes), which starts from freshly fetched brev-cli origin/main (a51d348a). This PR contains only start guidance changes in its review diff. BREV-13202 produces the WORKSPACE_ALREADY_EXISTS server code; deploy that server contract before expecting the coded guidance in released CLI builds. BREV-13207 separately updates the live brev create GPU retry path and should consume #515.

[patelspratik]

what's the consequence for older vesrions? they just get the wrong error message?