بنقرة واحدة
delegation
Spawn a sub-agent, or hand off to another profile window.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
القائمة
Spawn a sub-agent, or hand off to another profile window.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Hand a task to another agent window via send_to_window.
Search the live web for current information, facts, and documentation.
Browse repos, triage issues end-to-end, inspect PRs, and read files on GitHub.
Read, create, and update events on the user's Google Calendar.
Read, search, send, and label-manage Gmail messages for the user.
Browse and manage Linear issues, projects, teams, and workflow states.
| name | delegation |
| description | Spawn a sub-agent, or hand off to another profile window. |
| scope | agent |
Two delegation paths:
1. spawn_sub_agent — visual sub-agent. Use for (a) tasks substantial enough that inlining would block other work, (b) tasks independent of this conversation, or (c) naturally parallelisable work ("research these 3 companies"). Sub-agents see nothing from this conversation — include every piece of context they need in task. Give them a concrete deliverable, not a vague goal. Max 3 per user turn.
2. send_to_window — hand to another open profile window. When a slice of the task fits another profile's scope (Coder for repos, Scribe for docs, Comms for messaging). The user gates the handoff via a permission dialog. See the cross-window skill for full guidance.
When you lack a tool the user is asking for, do not spawn a sub-agent in this version. Tell the user which profile would handle this and ask them to open one. Once the right profile is open, use send_to_window to delegate.
Don't delegate when you can do it in one step yourself. Don't hop across more than 2 windows — the system enforces a 3-hop limit. When you do delegate, give the user one line on what's happening and why.
Do not spawn a sub-agent to retry a tool call you just failed. The sub-agent has the same tools and will hit the same wall. Sub-agents are for independent work — different inputs, different deliverables — not for redoing your own work in a fresh context.