Auto-update regression (2.1.255) broke launch on macOS Monterey with no warning or rollback path

What happened

On 2026-09-02, an unannounced Claude Code auto-update killed a running session mid-work on an Intel Mac mini running macOS Monterey (12), with no rollback option available at the time. This cost a full day of productivity and the loss of continuity for the account/session running on that machine.

Investigating the public CHANGELOG.md afterward turned up a very likely match:

v2.1.258: "Fixed Claude Code failing to launch on macOS 12 (Monterey) from 2.1.255 regression."

This strongly matches the incident's OS and symptom (Claude Code failing to launch at all), suggesting the actual cause was an unintentional regression introduced in v2.1.255 — not a deliberate platform/OS support decision — that shipped via auto-update with no advance compatibility check and no rollback mechanism for the affected machine.

Impact

• A full day of lost productivity.
• Loss of session/account continuity on the affected machine (no way to resume where the session left off once the binary could no longer launch).
• No advance warning that an update could break launching on a still-supported OS version.
• No rollback path once the regression landed.

Ask

1. Consider a pre-flight compatibility/launch check before an auto-update replaces a working binary, so a regression like this can't silently strand a machine.
2. Consider a rollback mechanism (revert to last-known-working build) when the newly-updated binary fails to launch.
3. Changelog entries have no date stamps, which made reconciling "when did this regression exist and for how long" after the fact needlessly difficult — dating changelog entries would help future incident reconciliation.

Evidence

• CHANGELOG.md, anthropics/claude-code — entries for v2.1.255 through v2.1.258.
• Affected OS: macOS 12 (Monterey), Intel Mac.

Filed based on my own review of the public changelog after the fact; happy to provide more detail if useful.