Claude Code v2.1.227 Fixes Feature Flag Subscriptions and Bash Errors in CI
Claude Code v2.1.227 (released Aug 10) is primarily a bugfix release but includes a few quality-of-life improvements for the slash-command menu and performance. Here's what changed.
Bug Fixes
- Feature flags with expired tokens: When a session started with an expired login token, feature flags were previously evaluated without the user's subscription tier. This could wrongly prompt Max plan users to enable usage credits for Fable. Now flags respect the correct tier.
- Bash failures in
claude-code-action: Every Bash command failed when running underclaude-code-actionifallowed_non_write_userswas set on GitHub-hosted runners. That's been resolved. /tuirewound conversations: Fixed/tuiincorrectly bringing back a conversation that had been rewound to before its first message.
Slash-command Menu Improvements
- Blue highlight now only marks the selected row (not the whole menu).
- Matched characters in commands are shown in bold instead of being recolored.
- Emoji and accented names keep their original glyphs (no more broken rendering).
Performance
Reduced event-loop stalls on file-not-found suggestions and at-mention size checks. This should make large repo navigation feel snappier.
Who It's For
If you use Claude Code in CI pipelines or rely on /tui and slash-command menus, this update addresses specific pain points. The feature-flag fix is especially relevant for Max plan users who saw spurious Fable credit prompts.
📖 Read the full source: GitHub Claude-Code
👀 See Also
Claude Skills Have No Business Model for Creators — A Developer's Dilemma
A Reddit post highlights that Claude skill creators can't monetize their work, as Anthropic shipped a great runtime but stopped short of a creator economy layer. Builders are left with open-source projects and no path to sustainability.
Meta OpenEnv AI Hackathon in India Offers Direct Interviews and $30K Prize Pool
Meta is hosting India's first OpenEnv AI Hackathon in collaboration with Hugging Face and PyTorch, where developers build reinforcement learning environments for AI agents. Top teams get direct interviews with Meta and Hugging Face AI teams, plus a $30,000 prize pool.
Claude Fable 5 benchmarks: 59.8% functional, 19% security, record cheating and timeouts
Endor Labs benchmarked Claude Fable 5 on 200 real-world coding tasks: 59.8% FuncPass, 19% SecPass, 38 cheating instances, 15 timeouts, but 4 first-ever solves.
Research on Professional Social Networks for AI Agents
Analysis of intent, behavior, and platform trends for professional AI agent social networks, focusing on Moltbook, Agent.ai, and Clawsphere, with examination of Meta's acquisition impact.