| name | main-brain-dispatch |
| description | Use when acting as main_brain inside a rendered MathorCup instance and you need to inspect the ready pool, dispatch a bounded task, observe its status, or decide close, reopen, or cancel through repo scripts instead of hand-editing runtime truth. |
Main Brain Dispatch
Use this skill for task orchestration in a rendered instance.
Read order
.codex/requirements.toml
AGENTS.md
MEMORY.md
project/spec/runtime_contract.md
project/spec/multi_agent_workflow_contract.md
Default entry chain
bash scripts/doctor.sh --target <dir>
bash scripts/main_brain_summary.sh --target <dir>
bash scripts/show_task.sh --task <task_id> --target <dir>
Dispatch path
bash scripts/list_open_tasks.sh --open-only --target <dir>
bash scripts/dispatch_task.sh --task <task_id> --owner <owner> --target <dir>
dispatch_task.sh is the canonical path for feedback skeleton creation.
dispatch_task.sh writes a packet artifact under project/workflow/packets/ by default.
submit_feedback.sh is only for repair or retrospective initialization.
- Prefer stage-level modeling tasks when possible:
TASK_CODE_MODEL_P1
TASK_CODE_MODEL_P23
TASK_CODE_MODEL_P4
- Code/model stages that produce canonical numbers or assumptions should maintain
project/output/model_manifest.json from project/output/MODEL_MANIFEST_TEMPLATE.json.
Acceptance helpers
bash scripts/paper_acceptance_check.sh --target <dir> --write-report
bash scripts/artifact_index.sh --target <dir>
- Run
paper_acceptance_check.sh after paper build and before final paper acceptance.
- Use
artifact_index.sh when packets, feedback, callbacks, exec runs, handoffs, and adjudications are scattered.
State transition rules
- Use
close_task.sh, reopen_task.sh, and cancel_task.sh.
- Do not hand-edit:
project/runtime/task_registry.json
project/runtime/work_queue.json
project/runtime/event_log.jsonl
Human gate
The main brain still decides acceptance. Native bridge files do not replace repo state or repo scripts.