Skip to content

Fleet spawn times out unconfirmed after worker is live and attachable #1838

Description

@kjgbot

Reproduction (2026-09-22, broker 12.4.0)

Run a targeted, confirmed agent-relay fleet spawn codex --node sf-frame --name relay-delivery-gap-repair --cwd /home/khaliqgant/Projects/AgentWorkforce/relay --task ... from a pinned workspace. The node acknowledges invocation inv_228071519860695040, but the CLI waits the full 120000 ms and exits spawn_unconfirmed / unconfirmed_may_be_running with no result.

During that same wait, agent-relay fleet agent list --json already listed relay-delivery-gap-repair as a live, node-owned worker on sf-frame. agent-relay node agent attach relay-delivery-gap-repair --node sf-frame --mode view succeeded and showed it executing the assigned task. Do not retry blindly: that would duplicate work.

This is a confirmed-success reporting gap, distinct from a genuinely accepted-but-never-launched spawn. It appears related to the previously closed #1563 but persists on 12.4.0.

Acceptance

  • A registered and attachable target worker yields a successful confirmed spawn result with spawned:true and ready:true proof (or equivalent), before the confirmation timeout.
  • A broker receipt lost after registration can be reconciled against the exact invocation/name/node identity without accepting a different worker as proof.
  • Deterministic integration coverage distinguishes live-and-ready, accepted-but-not-launched, and launched-but-unregistered; no false success and no false failure.
  • The CLI and MCP surfaces report the same reconciled outcome and preserve retry safety.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions