Parallel Dispatch Planning
Original, project-agnostic guidance. Agent-host effectiveness remains experimental.
When to use
Select genuinely ready parallel work without colliding on shared files, mutable resources, unstable contracts, or review capacity. Activate only when the current task crosses this boundary.
When not to use
Do not impose orchestration on a trivial edit or use these instructions to expand authority, start paid agents, alter production, or replace the consuming host’s policies.
Procedure
Verify readiness
Read the current graph and accepted dependencies. Submitted work is not accepted work; starting a child against an unreviewed upstream artifact can multiply rework. Keep decisions and scope blocks explicit.
Inspect collision domains
Check file and directory ownership, generated outputs, compiler state, ports, databases, queues, and shared provider quotas. A worktree only separates some source state. Account for running or indeterminate previous attempts before scheduling replacements.
Admit within a shared budget
Reserve tokens and resource slots for the proposed wave, plus capacity for verification/integration. Keep LLM worker count distinct from compiler and database parallelism. Use Agentflow schedule as an offline admission proposal, not an actual dispatcher.
Revalidate at the execution boundary
The private launcher must atomically reserve resources, confirm approvals and context freshness, and start only the selected bounded operation. Repeatedly running a planning command does not reserve anything or make its proposals unique.
Output
Selected tasks and workers, resource/token reservations, deferred tasks with reasons, and launch-time checks. Tie observations to actual inputs and keep passed, failed, blocked, and not-run states distinct.
Failure handling
An interrupted task may still be running on the same files: reconcile it before admitting a conflicting replacement. Preserve partial work and report the smallest missing prerequisite rather than manufacture evidence.
Example
Two independent adapters can run together, but the third implementation waits so review retains a slot. Their shared generated client has one upstream owner. This is a synthetic design scenario, not a completed host evaluation.
Companion tooling and evaluation
Agentflow commands and boundaries document the optional offline coordination helper. It performs only its documented checks; the full procedure still needs a qualified host and private adapter. Use references/scenarios.json for trigger, boundary, and non-trigger evaluation inputs. Their not-run status is not a test result.