TL;DR
sandbox.credentialsblocks credential files and secret env vars in sandboxed commands. If a sandboxed script suddenly can't find your AWS or GCP creds, this new setting is why.Workflow
agent({schema})structured output no longer loops or drops data. The model could re-callStructuredOutputindefinitely after a successful call, and follow-up turns sometimes failed to return structured output (more below).Remote MCP tool calls now abort after 5 minutes of silence instead of hanging forever. Override the timeout with
CLAUDE_CODE_MCP_TOOL_IDLE_TIMEOUTif your tools legitimately take longer.Subagent depth tracking fixed for resumed and forked agents. Resumed subagents restore their original spawn depth, and forks now count toward the 5-level cap, closing a gap where deep chains could exceed the limit.
New in 2.1.187
2.1.187 (June 24, 2026)
- Added
sandbox.credentialssetting to block sandboxed commands from reading credential files and secret environment variables - Added org-configured model restrictions to the model picker,
--model,/model, andANTHROPIC_MODEL, with a "restricted by your organization's settings" message when a restricted model is selected - Added mouse click support to select menus (permission prompts,
/model,/config, etc.) in fullscreen mode - Fixed
--resumefailing with "No conversation found" when the original-prun produced no model turns - Fixed
--json-schemaand workflowagent({schema})structured output: the model can no longer re-callStructuredOutputindefinitely after a successful call, and follow-up turns now reliably return structured output - Fixed remote MCP tool calls that hang with no response for 5 minutes, they now abort with an error instead of blocking indefinitely (override with
CLAUDE_CODE_MCP_TOOL_IDLE_TIMEOUT) - Fixed Claude Code Remote sessions taking ~2.7s longer to start after the agent proxy CA system-trust install was added
- Fixed pasted Korean/CJK text turning into mojibake in terminals that deliver paste as per-byte extended-key events
- Fixed
/updateover Remote Control hanging when a startup trust dialog would have shown - Fixed background jobs in the agents view getting stuck in "working" indefinitely when the agent ended a turn without producing structured output
- Fixed channel connections dropping after navigating to the agents view and back, and after
/bg,/tui, or/update - Fixed agent stop notifications not correctly attributing who stopped the agent, and improved wording ("finished"/"stopped" instead of "came to rest")
- Fixed subagent depth tracking: resumed subagents now restore their original spawn depth, and forked subagents now count toward the depth cap
- Fixed leaked agent worktree registrations: locked
.git/worktrees/entries from killed agents are now cleaned up automatically - Fixed Cmd+click not opening URLs in fullscreen mode in Ghostty on macOS
- Fixed
claude --helpnot listing the--bg/--backgroundflag - Fixed Esc, Ctrl-C, and Ctrl-D not working while
/shareis uploading - Improved
/install-github-app: GitHub Actions workflow setup is now optional, you can install just the GitHub App and skip the workflow/secret steps - Improved
/btwwith left/right arrow navigation to step through earlier answers - Improved
/pluginto surface plugins you haven't used recently so you can clean them up - [VSCode] Fixed extension becoming unresponsive when resuming a large session
Notes
Structured output fix completes the 2.1.186 arc
2.1.186 fixed workflow agent({schema}) subagents looping forever on repeated schema validation failures (aborting after 5 attempts). 2.1.187 fixes a different failure on the same surface: the model re-calling StructuredOutput after it already succeeded, and follow-up turns dropping structured data entirely. If you saw intermittent null returns or runaway agent loops in schema-driven workflows, both versions together close the gap.
Sandbox hardening continues
The new sandbox.credentials setting is the third sandbox tightening in two weeks. 2.1.183 blocked destructive git commands and infrastructure teardowns in auto mode. 2.1.181 added sandbox.allowAppleEvents as an opt-in gate. Now credential files and secret env vars are blocked by default. The trajectory is clear: sandboxed commands are converging on a minimal-privilege baseline where each category of access is an explicit opt-in.