Root ยท coordinate the taskGPT-6.1 Sol / mediumDelegate independent tasks when the benefits justify the coordination cost. Keep small or tightly coupled tasks with one agent.
SUBAGENTS / GPT-6.1 SOL
Define scope, deliverables, file ownership, context, and acceptance conditions. Require evidence and unresolved questions.
LowBounded tasks with explicit requirementsLittle interpretation required. Targeted searches, factual extraction, prescribed replacements, and mechanical checks. Examples: locate callers, extract toolchain versions, or check file equality.
MediumClear requirements with connected decisions or checksRoutine implementation, bounded research, test development, and debugging with an established failure. Examples: implement a defined feature or trace a known workflow.
HighUnresolved ambiguity or complex interactionsSubstantial analysis required. Unclear failure causes, competing hypotheses, consequential boundary changes, and difficult compatibility decisions. Examples: diagnose recovery races or trace account isolation failures.
Choose effort for each subtask. Use the lowest effort that meets its requirements. Read-only work can still require high effort.
Integrate & verifyRoot / GPT-6.1 Sol / mediumWait for all relevant edits to finish. Combine results, resolve conflicts, and run required checks. Keep source stable during verification and review.
Independent final reviewGPT-6.1 Sol / xhighAfter integration and required checks, inspect actual changes and affected consumers. Assess requirements, correctness, regressions, and verification evidence.
The reviewer stays independent of implementation. The root resolves confirmed findings and repeats affected checks. The reviewer checks corrections before completion.