Claude Code dynamic workflows Image: Anthropic / claude.com
by VibecodedThis

Claude Code 2.1.202 Adds a Dynamic Workflow Size Setting and Workflow Telemetry

Claude Code 2.1.202 ships a /config knob for controlling how large dynamic workflows grow, OpenTelemetry attributes for correlating workflow agent activity, a cleaner /workflows UI, and fixes for Remote Control drops, session slowness in multi-worktree repos, and more.

Share

Claude Code 2.1.202 released today with one new setting, one new telemetry feature, and a long list of bug fixes.

Control dynamic workflow size

The headline addition is a “Dynamic workflow size” option in /config. It lets you tell Claude how large to make its dynamic workflows: small, medium, or large agent counts. The setting is an advisory guideline, not an enforced cap, but it gives teams a way to dial down the footprint when cost or concurrency limits are a concern, or dial up when a task needs maximum parallelism.

Dynamic workflows, which Claude introduced in May, let a single session spin up tens or hundreds of parallel subagents to tackle large tasks end to end. The new size setting gives you a lever without having to prompt-engineer around Claude’s default judgment.

Workflow telemetry via OpenTelemetry

Also new: workflow.run_id and workflow.name OpenTelemetry attributes on workflow-spawned agents. If you’re collecting traces from Claude Code, you can now correlate individual agent activities back to the workflow run that spawned them. Useful for debugging multi-agent sessions or building dashboards over agent behavior in CI.

Cleaner /workflows view

The /workflows agent list got a layout pass: wider title column, a dedicated time column, and shorter model name display. Per-row tool-call counts were removed. The result is easier to scan when you have several agents running simultaneously.

Bug fixes

A few fixes worth calling out:

Remote Control: Images and files sent from Remote Control apps without a caption were being silently dropped. That’s fixed. Separately, Remote Control commands were sometimes failing with an “Unknown command” error; that’s also fixed. And the sign-in URL format has been corrected so it’s reliably clickable when you’re connecting over SSH.

Session resumption in multi-worktree repos: Resuming a session in a repository with many git worktrees could cause multi-minute delays and excessive memory use. Fixed.

History search crash: Accepting or cancelling an inline Ctrl+R history search while a file scan was still running caused a crash. Fixed.

Background session rename: Using /rename on a background session was reverting to the previous name when the session’s jobs restarted. Fixed.

Skills: Re-invoking a skill was duplicating its instructions in the context window. Fixed.

Voice input: A microphone failure was triggering an unbounded retry loop. It now pauses instead of spinning.

Proxy/network: A transient connection drop during a download was immediately failing rather than retrying. The download now retries automatically on transient failures.

/review command: The /review <pr> command had been changed to a multi-agent operation in a prior release. It’s been restored as a fast, single-pass review. If you want the multi-agent version, use /code-review.

MCP: Error messages for misconfigured MCP server types have been improved to give clearer guidance on what’s wrong.


Full changelog: code.claude.com/docs/en/changelog

Share