Claude Code desktop app interface showing an active coding session Anthropic
by VibecodedThis

Claude Code v2.1.145-146: `agents --json` Scripting and the `/code-review` Rename

Two back-to-back Claude Code releases ship a JSON flag for listing live sessions, improved plugin discovery, better OpenTelemetry tracing, and a rename of /simplify to /code-review with optional effort levels.

Share

Anthropic shipped two consecutive Claude Code releases over the past three days. v2.1.145 landed May 19, v2.1.146 on May 21. Together they add a handful of developer-quality-of-life improvements alongside the usual round of bug fixes.

claude agents --json

The standout addition in v2.1.145 is a --json flag for claude agents. It lists all live Claude Code sessions as structured JSON, which makes it straightforward to wire up external tooling without scraping terminal output.

The use cases are practical: feeding session state into a tmux-resurrect config, surfacing active agents in a shell status bar, or building a session picker script. Before this flag, none of that was clean to automate. Now you pipe the JSON and parse what you need.

The same release added agent_id and parent_agent_id attributes to OpenTelemetry spans on claude_code.tool. Trace parenting is also fixed: background subagent spans now nest correctly under the dispatching Agent tool span rather than appearing at the root level. If you’re correlating Claude Code activity with application observability, this makes the traces actually readable.

The /plugin Discover and Browse screens got a layout update in v2.1.145 as well. They now show a plugin’s commands, agents, skills, hooks, and connected MCP/LSP servers together in one view — useful when evaluating a plugin before you commit to installing it.

The terminal tab title for claude agents also now shows the count of sessions waiting on your input. If you’ve got an alt-tabbed window running background agents, you’ll know when something needs attention without switching to it.

/simplify is now /code-review

v2.1.146 renamed the /simplify command to /code-review. The command accepts an optional effort level: /code-review high tells Claude to spend more time on the pass.

The rename reflects what the command actually does more accurately. A proper code review catches style problems, redundant logic, and missed edge cases — simplification is part of it, but not the whole picture. The old name undersold the scope.

Bug fixes worth noting

v2.1.146 clears several platform-specific issues that had been causing friction:

  • Windows PowerShell: Users with pwsh installed via winget or the Microsoft Store were hitting a “command line is invalid” error on tool calls. Fixed.
  • MCP pagination: Paginating MCP servers were silently dropping resources and prompts past page 1. Fixed across resources/list, resources/templates/list, and prompts/list.
  • Windows Terminal strobing: Full-screen mode was producing strobing artifacts in background sessions. Fixed.
  • Permission re-prompts: Backgrounded sessions were asking for tool permissions that had already been granted in the parent session. Fixed.
  • Subagent model forwarding: CLAUDE_CODE_SUBAGENT_MODEL wasn’t propagating correctly to child processes. Fixed.
  • Network reliability: The auto-updater now retries on transient failures instead of giving up on the first network error.

v2.1.145 also patched a permission-prompt bypass affecting bare environment variable assignments in Bash, which is worth noting for anyone running Claude Code in security-sensitive environments.

The full release notes for v2.1.145 and v2.1.146 are on GitHub.

Share