Skip to content

Research task-level orchestration model for Buran #2

@MrFlashAccount

Description

@MrFlashAccount

Problem: flat fan-out from one Buran run into many role workers does not scale and wastes context.

Target model: Buran run -> task-level agents -> DevHarness role workers.

Buran should spawn one task-level subagent/orchestrator per task/issue/slice; each task agent runs implementer/reviewer/critic/architect/QA workers as needed.

Default workspace ownership should be task-scoped; role-scoped workspaces only for competing implementations or explicit isolation needs.

Status: future research/implementation, not current behavior.

Research questions:

  • OpenClaw nested subagent limits / maxSpawnDepth
  • DevHarness integration boundary
  • recovery / cancellation semantics
  • per-task state ownership
  • workspace lifecycle
  • observability / logging
  • fan-out safeguards

Acceptance criteria for future implementation/research doc:

  • clarifies the orchestration boundary and depth limits
  • defines task-scoped workspace ownership and lifecycle
  • defines cancellation / recovery behavior
  • defines logging / observability requirements
  • identifies safeguards against unbounded fan-out
  • notes this is future work, not current runtime behavior.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions