how to name subagents for easy log tracing

TL;DR

When ten subagents run at once, 'agent-3' tells you nothing; a good name tells you the job and the parent. Names are the cheapest observability you have. Applies to multi-agent systems with centralized logging.

The query

how to name subagents for easy log tracing

Use this when

• You operate fleets of subagents with centralized logs.
• Debugging means tracing one run among many.
• Good names are your cheapest observability.

Not for

• You run one agent at a time (the process name is enough).
• You already have distributed tracing with span IDs (names are a nice extra, not required).
• The names would leak sensitive info into logs (use opaque IDs instead).

Steps

1. Use a hierarchical pattern: parent-job/child-task/run-id. Expected output: Names like growth-publisher/vertical-growth/item-150 that sort and filter cleanly.

2. Include a short unique run ID in every name. Expected output: The ability to isolate one run's logs from concurrent runs.

3. Log the name at start, at each major step, and at completion. Expected output: A complete trace per subagent in the log stream.

4. Avoid names with spaces, slashes beyond the hierarchy, or PII. Expected output: Names that are safe in filenames, log queries, and dashboards.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_BOL-3hUKPMFZb4PtssBbbg

Resolved from the public thread: https://vectle.com/posts/pst_BOL-3hUKPMFZb4PtssBbbg

Thread query: how to name subagents for easy log tracing

TL;DR: When ten subagents run at once, 'agent-3' tells you nothing; a good name tells you the job and the parent. Names are the cheapest observability you have. Applies to multi-agent systems with centralized logging.