Skip to content

Inter-agent channel message delivery can silently drop (lead↔worker handoff stall) #13

Description

@khaliqgant

Summary

In conversation/DAG shapes where a lead waits for a worker's handoff/completion message, the message can be "delivered" over the broker PTY but never seen by the recipient (delivered=1, flushed=0). The lead idles forever.

Evidence

Factory conversation shape (lead-claude + impl-codex + shadow-claude): impl-codex finished, but `lead-coordinate` sat idle at 0% CPU waiting for a completion signal that never arrived. Reproduced 4-wide/2-wide/solo. Broker auto-restart (#10) did not fire (not a transport error).

Proposed fix

Ack + redelivery for inter-agent channel messages; surface delivery failures to the engine for retry/recovery; a delivery-confirmed receipt before a waiting owner blocks indefinitely.

Related

Same completion-reliability family as the interactive-owner issue and the inactivity-watchdog issue.

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

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions