dispatch

name	dispatch
description	Dispatch tasks to high-performance subagents. Use when asked to investigate, implement, audit, or delegate work to subagents.

Dispatch

How to delegate tasks to frontier subagents.

Defaults

• Investigate / audit / research: dispatch two agents in parallel — one Claude (opus series, highest available) and one OpenAI (GPT sol series, highest available). Read-only — forbid edits.
• Implement / fix / build: dispatch one Claude agent (opus series). Allow edits, forbid commits.

Always use long_context context tier and the highest reasoning effort available for the model.

Seed prompt rules

1. Explain enough context that the agent can work without asking.
2. State the user's exact instructions — what to do, what to report.
3. If the agent needs prior conversation context, summarize it in the prompt — agents are stateless.

Model selection

Pick the largest number in each series. Examples:

• Anthropic: claude-opus-5, claude-opus-4.8, claude-opus-4.7
• OpenAI: gpt-5.6-sol, gpt-5.6-terra, gpt-5.5

These are examples — versions change. If unsure which models are available, look up the available models list or check the Copilot CLI documentation.

Agent type

• Needs to search GitHub / web / external repos → research
• Needs to read, grep, run commands in this repo → general-purpose
• Needs to critique a plan or implementation → rubber-duck

Constraints

• Investigating agents must not edit files or run destructive commands.
• Implementing agents must not commit. The user reviews and commits.
• Always use mode: "background" and report results when agents complete.