Claude Code ships Claude Mods to stable, plus a watchdog side agent

Version 2.1.287 turns behavior-rewriting plugins on by default — and quietly adds MCP sign-in prompts, a built-in gh api, and a restored rm safeguard.

Nowline OCT 2 7:00 AM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • Claude Mods graduate to stable

    Mods — plugins that hook engine events to rewrite Claude's behavior mid-run, not just bolt on instructions — ship in 2.1.287 after living behind the CLAUDE_CODE_ENABLE_FUNCTION_HOOKS flag since early September. You can now distribute plugins that modify deeper behavior to anyone on the stable channel.

  • 'You should know' watches your back

    A new built-in mod runs a side agent that flags things you or Claude might miss during a session. Turn it on with /plugin enable cc-plugin-you-should-know@builtin (first-party sessions with telemetry on). It's the first shipped Mod that does real work rather than just demoing the API.

  • MCP servers can prompt you to sign in

    Servers on the 2025-11-25 protocol can now push URL prompts — an OAuth sign-in link, say — straight into your session. If a server stops connecting after the update, add "bareElicitationCapability": true to its MCP config entry.

  • Built-in gh api for self-hosted runners

    Sessions using Anthropic-managed git on macOS and Linux now get a built-in gh api (REST only) on machines where the GitHub CLI isn't installed — one less dependency to provision on your own runners.

  • A dangerous rm gets its guardrail back

    2.1.287 patches a hole where a destructive rm (on / or your home directory) lost its always-ask safeguard if the same command also redirected output to a ~ or wildcard path. Worth updating for this fix alone.