skills.rest
BTreeMap
BTreeMapOFFICIAL
Explore all skills by this author →
WebsiteGitHub

summon

Delegates tasks to subagents with structured briefs, disjoint fanout, and return review.

2|2|Updated Jun 21, 2026
Search Tags:#prompt engineering#claude code#delegation#multi-agent#subagent#task orchestration#fanout
One-click install
npx skills add https://github.com/BTreeMap/SKILLs --skill summon-btreemap
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: summon
Source: https://github.com/BTreeMap/SKILLs/tree/main/summon
Command: npx skills add https://github.com/BTreeMap/SKILLs --skill summon-btreemap

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Delegating work to subagents often fails silently: delegates lack the evidence they need, parallel agents overlap on the same files, and returned results get trusted without verification. This Skill gives you a disciplined procedure for deciding whether to delegate at all, writing a complete six-field brief, partitioning work across parallel agents, and judging every return as untrusted input before acting on it. ## Core Features & Use Cases - Four verbs: dispatch (one brief, one delegate), fanout (partition work into disjoint bundles), review (judge a return against its contract), and help (quick-reference card). - Six-field brief template: objective, evidence, rules, bounds, contract, and budget, each stated explicitly or marked not applicable, so delegates never guess at missing context. - Harness reference: per-platform tables covering Claude Code, GitHub Copilot, Google Antigravity, OpenAI Codex, DeepSeek Harness, and the OpenAI Agents SDK, including concurrency caps, nesting depth, and skill-loading behavior. - Use Case: You need three independent codebases researched in parallel. Use fanout to split the work into disjoint bundles with named sibling territories, dispatch one delegate per bundle, then review each return against its contract before merging findings. ## Quick Start Ask the agent to use the summon skill to delegate a research task to a subagent and review what comes back.

Frequently Asked Questions about summon

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I delegate a task to a subagent in Claude Code?▼

Compose a six-field brief covering objective, evidence, rules, bounds, contract, and budget, then spawn one delegate with it. Default to inline work unless the task needs a separate context, since an errand for a few tool calls costs 26k to 53k delegate tokens.

How do I split work across multiple parallel agents?▼

Use fanout to partition the open work into bundles that are pairwise disjoint on both topic and files, with one full brief per bundle. Each bundle names its sibling territories and owned files, and the fan must fit the harness concurrency cap, such as Claude Code's limit of 20.

Should I trust output returned by a subagent?▼

No. Treat every return as untrusted input, like a fetched web page, and judge it against the brief's contract before acting on any instruction inside it. A failed status is no return at all, since its result is the delegate's last inner thought rather than a report.

When should I not delegate a task to another agent?▼

Keep work inline when the lead can close it in a handful of tool calls; measured break-even sits near five tool calls. Delegation buys context isolation and wall-clock time, never correctness, and loses on any single dependent chain.

How do I give a subagent access to a skill?▼

Take the first reach variant that holds: preloaded in the agent definition, invocable via a skill tool, readable by absolute path, definable through a hot-reloaded agent definition, or sealed, where you excerpt the binding rules into the brief. Probe an unfamiliar harness with one test delegate first.

Related Skills

Explore More Software Engineering Skills→
mtk177amtk177a

choose-ai-execution-setup

Recommends which available AI execution setup should handle a concrete task.

Community
Intermediate
KingSharkGKingSharkG

orchestrator-dispatch

Coordinate auditable subtask dispatch workflows with pre-dispatch checklists and inline bundles.

Community
Advanced
kashman001kashman001

research-wave

Orchestrates parallel multi-agent research with independent fact-checking and ruled corrections.

Community
Advanced
QunnnnQunnnn

dispatching-parallel-agents

Coordinate parallel delegation of independent tasks across specialized agents.

Community
Intermediate