Claude Code 2.1.161 makes agent work measurable by team

Today's Claude signal is for teams that want to use agents as real work, not just as experiments in a side tab. Claude Code 2.1.161 now includes OTEL_RESOURCE_ATTRIBUTES values as labels on metric data points. Agent runs can be sliced by things like team, repo, or workflow instead of disappearing into one anonymous AI cost bucket.

Source: Claude Code changelog

Why this matters in everyday operations

OpenTelemetry is a standard for sending metrics, logs, and traces to an observability system. Claude Code can already export usage, costs, tool activity, events, and traces through OpenTelemetry. The 2.1.161 change is that resource attributes now travel with the metric data points as labels.

Source: Monitoring Claude Code usage

That sounds technical, but the operational question is plain: which team is getting value from Claude Code, which repo is burning the most agent time, and where do runs stall before a human can review the result? Without labels, the answer often turns into guesswork and screenshots.

Two small changes make runs easier to follow

The same release also changes how parallel tool calls behave. If one Bash command fails inside a batch of parallel tool calls, it no longer cancels the other calls. Each tool returns its own result independently. claude agents also shows done/total when work has been fanned out, and peek highlights the longest-running item.

Source: Claude Code changelog

For small teams, the point is not to measure everything. It is to stop treating agent work as magic. An agent run should be able to leave a simple receipt: which workflow it belonged to, which repo it touched, which tools failed, and what still needs human review.

Try this prompt this week

Do the human part first: choose a bounded workflow, decide which labels you actually need, and keep them free from personal data and secrets. Names like team, repo, and workflow go a long way. If you run Claude Code centrally, those environment variables can be managed through administered settings in Anthropic's documentation.

Then ask Claude to help with the routine:

Read our Claude Code settings, telemetry runbook, and docs for [workflow].
Suggest 3-5 OTEL_RESOURCE_ATTRIBUTES that help us track team, repo, and workflow without personal data.
Show which Claude Code events or tool failures should appear in a weekly run receipt.
Do not change anything yet. First give me a short plan a human can approve.

This is a Tool Forge problem. Not whether you should use Claude, but how Claude work becomes traceable enough to use more often. Start with one repo and one recurring workflow. When the receipt is readable, connect more.