Claude Code Digest — 2026-04-07 20:16:47
What the docs reveal
Taking Control of Timeouts
Anthropic exposed granular timeout variables for API requests, bash execution, and Model Context Protocol (MCP) tools. This signals a fundamental shift: Claude Code manages increasingly complex, long-running local workloads, and hard-coded timeouts no longer suffice.
Developers operating high-latency networks or running heavy MCP integrations previously hit silent ceilings. By introducing API_TIMEOUT_MS, BASH_MAX_TIMEOUT_MS, and MCP_TOOL_TIMEOUT, Anthropic hands execution control back to operators. Tune these variables to explicitly accommodate slow internal microservices or massive compilation scripts. The era of arbitrary command truncation is over.
Hook Exit Codes: A Crucial Distinction
The documentation clarifies a critical behavioral change in custom hooks. Claude Code now treats exit 1 as non-blocking, while exit 2 explicitly blocks the current action.
Anthropic made this engineering trade-off to solve a widespread workflow breakdown. Standard linting and testing utilities frequently return exit 1 for minor formatting warnings. Previously, these soft failures abruptly halted the entire agent loop. Now, Claude gracefully logs the failure and proceeds. You must consciously issue exit 2 to stop an AI action. The single exception is WorktreeCreate, which still aborts on any non-zero exit code to prevent corrupting local git state. Audit your custom hooks immediately. If a local security check must stop Claude Code, rewrite that script to exit with code 2.
Taming Model Misconfiguration
Anthropic expanded the troubleshooting guide to address "Model not found" errors and clarified the configuration priority stack. The underlying cause of this friction is model churn. As Anthropic deprecates legacy models and releases new variants, developers blindly copy stale configuration files between projects. This practice traps agents in broken states.
To solve this, Anthropic introduced the interactive /model command. This feature proves developers demand modern CLI ergonomics over manual string patching for model discovery. Stop memorizing cryptic API strings. Clear your legacy environment variables entirely, run the /model command, and let the software handle state management.
Debugging the VS Code Integration
The VS Code documentation now features a dedicated troubleshooting section detailing local extension verification and log access. Operating an autonomous agent within a highly customized IDE invites friction. Shell path conflicts, process timeouts, and extension host crashes are inevitable side effects of tight editor integration.
By defining exact log locations, Anthropic shifts debugging capability directly to the user. When the VS Code extension inevitably hangs during a complex refactor, skip the public issue tracker. Read your local system logs first.