Claude Code Digest — 2026-03-16 20:15:06
What the docs reveal
Anthropic is tackling prompt fatigue. Constant permission prompts protect local environments, but they stall automated workflows. New documentation details PermissionRequest hooks. Developers can now programmatically automate standard prompt responses by writing JSON decisions directly to stdout. You can auto-approve specific tools, update session states to acceptEdits, and apply "always allow" rules dynamically.
This update shifts Claude Code from an interactive assistant toward an autonomous agent. You define the guardrails in code. But Anthropic anticipates enterprise blowback. Server-managed settings now include a disableBypassPermissionsMode flag. Corporate IT can restrict developers from bypassing prompts entirely. Anthropic also tightly sandboxed third-party code. External plugin subagents systematically ignore hooks, mcpServers, and permissionMode configurations. If you want an agent to possess elevated permissions, you must define it locally in your project or user settings. Security trumps marketplace convenience.
Context loss kills productivity. Previously, updating a plugin or modifying an MCP server required restarting the application. Restarts wipe the context window. Anthropic solved this with the /reload-plugins command. Claude Code now hot-reloads commands, skills, agents, hooks, and MCP environments. The CLI outputs a direct summary of affected components and traps initializations errors. Claude Code also now initializes configured MCP servers automatically on startup. The application abstracts the lifecycle management. State persists.
Corporate networks break developer tools. Strict proxies and custom certificates routinely intercept and fail Anthropic's organization status checks. The new CLAUDE_CODE_SKIP_FAST_MODE_NETWORK_ERRORS environment variable acknowledges this reality. You can force fast mode to execute even when network blocks prevent upstream validation.
Anthropic restructured telemetry collection. Session quality surveys now run by default across all infrastructure providers, including AWS Bedrock, GCP Vertex, and Foundry. Anthropic needs varied performance data to assess model routing. Recognizing enterprise data privacy mandates, they provided new silencers. You can set the new feedbackSurveyRate setting to 0, pass CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY, or trigger a strict network cutoff with CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC. The latter drops all survey and telemetry pings at the edge.
Finally, the status line progress bar replaced the ▓ rendering with a solid █. The standard block character renders consistently across common terminal emulators.