Claude Code v2.1.215: /verify and /code-review No Longer Automatic

Claude Code v2.1.215 changes how the /verify and /code-review skills behave. Previously, the agent ran these automatically during certain operations. Starting with this release, they no longer run on their own — you must explicitly invoke them with the /verify or /code-review commands.
Key Changes
- Automatic execution removed: Claude will no longer trigger
/verifyor/code-reviewunprompted. - Manual invocation required: Use
/verifyor/code-reviewdirectly in the chat when you want those skills to run.
Impact on Workflow
This change gives you tighter control over when verification and code review happen. If you relied on the automatic behavior, you'll need to add explicit commands to your prompts or scripts. For users who found auto-verification disruptive, this is a welcome reduction in noise.
Updating
If you're on an older version, update Claude Code to v2.1.215 to apply this change. You can check your version and update via your package manager or download from the releases page.
📖 Read the full source: GitHub Claude-Code
👀 See Also

Claude Projects File Upload and Indexing Issues Documented
Claude Projects has multiple confirmed backend issues: files get stuck indexing, RAG search mode activates prematurely at ~13 files regardless of token count, and cached content persists even after deletion and re-upload.

sseanliu/VisionClaw Brings Real-Time AI Assistance to Meta Ray-Ban Smart Glasses
sseanliu's VisionClaw offers a revolutionary AI assistant for Meta Ray-Ban smart glasses, combining voice, vision, and agentic actions powered by Gemini Live and OpenClaw.

Go Players Disempower Themselves to AI: How Cheating Became Undetectable
The LessWrong post details how AI cheating in Go tournaments became rampant and nearly impossible to punish, using the case of Carlo Metta who used Leela 0.11 and Leela Zero to win 25 of 26 games over several seasons, with only one loss under camera surveillance.

Reddit user compares Claude Sonnet 4.6 and GPT-5 on 10 blogging tasks
A Reddit user tested Claude Sonnet 4.6 against GPT-5 using identical prompts for 10 common blogging tasks, finding the editing time difference to be the most useful metric.