We test a lot before releasing, but bugs still happen (for now). Which terminal are you using? Is there a specific action or tool call that tends to trigger scrolling to the top?
@bcherny
-
Claude Can Edit Its Own Settings Upon Request
By
–
You can ask it not to, and to add that to your settings. Claude is able to edit its own settings if you ask it
-
Claude Unshipped RAG for Search: Agentic Approach Better
By
–
We did RAG early on then unshipped it for a number of reasons: privacy, security, reliability, index staleness. Overall, we found that agentic search gave better results with fewer tradeoffs. If you prefer RAG, we recommend an MCP like Souregraph, or just ask Claude to build a
-
Verbose Mode Token Usage Impact Analysis
By
–
Verbose mode doesn’t impact token usage. Maybe you were just using it for a particularly token-intensive task?
-
User reports stable Opus 4.6 performance with 1m context window
By
–
I exclusively use 1m, haven’t noticed issues with opus 4.6
-
Claude verbose mode and contextual information display features
By
–
Ask claude to enable verbose mode if you want to see more. If a lot of people want more info for perplexity, we could also show details by default for it or make it configurable. As conversations get longer, we want to give you the info you need while hiding info you don’t need
-
Claude’s 1M Context Window: Users May Not Need Extended
By
–
With 1m context window, most users don’t need it anymore. If you miss it, ask claude to re-enable it in your settings.
-
Test Time Compute and Subagents: Optimizing AI Results
By
–
Roughly, the more tokens you throw at a coding problem, the better the result is. We call this test time compute. One way to make the result even better is to use separate context windows. This is what makes subagents work, and also why one agent can cause bugs and another
-
Model Effort Levels: Medium Balances Speed and Intelligence
By
–
Medium is a better default for most tasks. Think of effort as “max effort” — the model may decide to use less effort if it wants, and can use up to the amount of effort you set it to. It’s a balance- Low: fast and less intelligent
Medium: balanced
High: very intelligent, can be