Problem
When running a multi-agent crew pipeline (like the IMC 8-expert reviewer), different steps have very different execution times. A related-work-expert finishes in 60s, but a meta-reviewer that synthesizes all prior reviews can take 10+ minutes.
Currently, defaultTimeoutMs is set at the runtime level and applies uniformly to all steps. If you set it high enough for the meta-reviewer (600s), every step gets 600s timeout, which means failures take forever to surface. If you set it low (120s), the meta-reviewer times out.
Suggested fix
Allow per-step timeout override in the step builder:
step("meta-review")
.system("You are an area chair...")
.prompt("Synthesize all reviews: {reviewer-1} {reviewer-2}...")
.timeout(600_000) // 10 minutes for this step only
The pipeline executor should use step.timeoutMs ?? runtime.defaultTimeoutMs.
Real-world context
Discovered during the IMC 2026 paper crew review pipeline. The meta-reviewer step timed out repeatedly at 120s default, requiring a global timeout increase to 600s that slowed down failure detection for all other steps.
Labels
enhancement, crew
Problem
When running a multi-agent crew pipeline (like the IMC 8-expert reviewer), different steps have very different execution times. A
related-work-expertfinishes in 60s, but ameta-reviewerthat synthesizes all prior reviews can take 10+ minutes.Currently,
defaultTimeoutMsis set at the runtime level and applies uniformly to all steps. If you set it high enough for the meta-reviewer (600s), every step gets 600s timeout, which means failures take forever to surface. If you set it low (120s), the meta-reviewer times out.Suggested fix
Allow per-step timeout override in the step builder:
The pipeline executor should use
step.timeoutMs ?? runtime.defaultTimeoutMs.Real-world context
Discovered during the IMC 2026 paper crew review pipeline. The meta-reviewer step timed out repeatedly at 120s default, requiring a global timeout increase to 600s that slowed down failure detection for all other steps.
Labels
enhancement, crew