Sometimes it’s not the big features but the small annoyances that ruin your day. Claude Code 2.1.227, released on August 10, goes after exactly that kind of thing.
The bug that stung
When a session started with an expired login token, Claude Code evaluated its feature flags without knowing the user’s subscription tier. The result: Max plan users got prompted to enable usage credits for Fable – something their plan doesn’t require. A prompt that should never have reached them showed up anyway. It’s the kind of bug that quietly erodes trust, because for a second you think: wait, do I have to pay now?
The fix makes sure the subscription tier is read correctly at startup, before any flags kick in.
What else made the cut
If you run Claude Code in CI, the second fix matters more: under claude-code-action with allowed_non_write_users, every Bash command was failing on GitHub-hosted runners. That breaks pipelines, and it does so silently. Fixed.
There’s also one detail that has caught me out before: if you rewound a conversation with /tui, it could get pulled back up – including the messages from before the rewind. That’s clean now too.
The rest is polish. The slash-command menu now marks only the selected row in blue instead of recoloring matched characters – those get bolded instead. Emojis and accented names keep their glyphs. And performance: fewer event-loop stalls on file-not-found suggestions and at-mention size checks.
My take
None of this belongs on a keynote slide. But releases like this decide whether a tool grates on you day to day or just fades into the background – in the good way. The Fable credits prompt is a neat example of how a technical slip in token handling lands right on your wallet and generates stress out of nowhere.
Anthropic is shipping versions on a near-daily cadence right now. If you use Claude Code in production, update – especially for that CI fix.
Sources: