Summary
Calling a workflow-reference tool (a type: reference tool that runs another workflow server-side) fails with a timeout, and traces show the referenced workflow never started. The timeout happens before dispatch, not during a slow run.
Environment
- Harness: claude (Claude Code). Runner: sidecar. Sandbox: local.
- Calling agent: Chief of Staff. Tool: a workflow-reference tool ("Product_Agent" on that agent).
- Resolved variant slug: new-agent-ms4g9hpzo7.
- 2026-07-29 ~13:30 UTC.
Observed behavior
Two calls both failed with:
tool call workflow.variant.new-agent-ms4g9hpzo7 failed: The operation was aborted due to timeout
- query_spans shows zero spans after 13:27 UTC, so the referenced workflow never ran.
- The timeout occurs before any span is created, i.e. before dispatch.
- Verified no side effect occurred (no GitHub issue was created).
Interpretation (low confidence)
The slug "new-agent-ms4g9hpzo7" is NOT stale or wrong. It is the target agent's own variant slug (confirmed when that agent later committed a revision to workflow_slug "new-agent-ms4g9hpzo7"). So the reference resolved to the correct target. The timeout is pre-dispatch (zero spans created), so the hang is in the resolve/dispatch path itself, not in variant selection and not in a slow target run.
Reproduction (proposed, needs confirmation)
- On an agent that has a workflow-reference tool pointing at another workflow variant, invoke that tool.
- Observe: the call aborts with "operation was aborted due to timeout"; no span is created for the referenced workflow.
Impact
Agent-to-agent handoffs via workflow-reference tools are unreliable: the call can time out before the target workflow ever starts, with no partial execution and a generic timeout message that does not indicate whether the target resolved.
Related issues (not exact duplicates)
Asks
- Investigate the resolve/dispatch path for reference tools for a pre-dispatch hang.
- Surface a clearer error that distinguishes "variant not resolved" from "target ran but was slow."
Summary
Calling a workflow-reference tool (a
type: referencetool that runs another workflow server-side) fails with a timeout, and traces show the referenced workflow never started. The timeout happens before dispatch, not during a slow run.Environment
Observed behavior
Two calls both failed with:
Interpretation (low confidence)
The slug "new-agent-ms4g9hpzo7" is NOT stale or wrong. It is the target agent's own variant slug (confirmed when that agent later committed a revision to workflow_slug "new-agent-ms4g9hpzo7"). So the reference resolved to the correct target. The timeout is pre-dispatch (zero spans created), so the hang is in the resolve/dispatch path itself, not in variant selection and not in a slow target run.
Reproduction (proposed, needs confirmation)
Impact
Agent-to-agent handoffs via workflow-reference tools are unreliable: the call can time out before the target workflow ever starts, with no partial execution and a generic timeout message that does not indicate whether the target resolved.
Related issues (not exact duplicates)
Asks