Claude Code Digest — 2026-05-17 00:36:10
What the docs reveal
Enterprise-First Azure Foundry Configuration
Claude Code now connects to your Azure Foundry resources automatically upon the first prompt. The system demands environment variables like CLAUDE_CODE_USE_FOUNDRY and abandons the interactive setup wizard entirely.
Anthropic built this for enterprise workflows. Interactive CLI wizards break automated environments, CI/CD pipelines, and zero-trust bootstrapping. By forcing environment variables, Anthropic forces developers to manage credentials through standard infrastructure-as-code practices like .env files, direnv, or Azure Key Vault. You can no longer rely on a friendly terminal prompt to guide you through API key insertion.
Set up your .envrc now. If you move between multiple Azure Foundry deployments, script your environment variable swaps to maintain speed.
The Economics of Persistent Effort Levels
You can finally configure a default effortLevel directly in your settings file. Supported values include low, medium, high, and xhigh. The system outright rejects max as a persistent default, restricting it to session-specific commands.
This restriction protects your wallet. The max effort level leverages extensive reasoning loops and consumes tokens rapidly. If developers could save max in their global or project configurations, they would routinely burn API credits on trivial tasks like fixing typos or renaming variables. By requiring an explicit command for the maximum tier, Anthropic ensures you only pay for heavy compute when you deliberately request it.
Configure your standard project baseline to medium or high. Reserve the /effort max command for systemic architectural refactors or deep debugging sessions.
Sandboxing Skill Execution
Command substitution in Claude Code skills now runs exactly once per file. The documentation explicitly bans nested or recursive placeholder expansion.
This limitation prevents infinite execution loops. If templates could expand recursively, a placeholder evaluating to another placeholder would freeze the agent. More importantly, single-pass expansion prevents prompt injection vulnerabilities. If an external API returns a string matching a placeholder format, Claude Code will ignore it instead of executing unverified commands. Anthropic trades template flexibility for execution stability.
Stop trying to build complex, multi-stage pipelines inside placeholder templates. If your custom skill demands recursive logic, write a dedicated bash script or Python utility and instruct Claude to run it. Keep your templates flat, simple, and deterministic.