How do I install the Agent Delegation skill?

Run npx skillmds@latest add jezweb/agent-delegation in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Agent Delegation safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Agent Delegation?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Agent Delegation free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Agent Delegation?

jezweb (@jezweb) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Agent Delegation by jezweb — Agent Skill

Name: Agent Delegation
Author: jezweb

Agent Delegation

Use when about to delegate a discrete task to a sub-agent, deciding whether to spawn one at all, how much to specify it, and how to keep its work from bloating the main thread. The one-off-task layer below run-a-role-agent. Triggers, "delegate this", "spawn a sub-agent", "should this be a sub-agent or a script", "fan this out".

jezweb 25e1f03 2.3 KB Updated Sep 17, 2026

File contents

Agent-delegation: hand off a task well

Before you spawn a sub-agent, a couple of decisions decide whether the hand-off helps or just adds cost and flakiness.

First: should this be an agent at all, or a script? If the work can be described as code, write the code. Agents are for judgment; scripts are for precision. Copying files, transforming a format, a deterministic sweep, those want a script, not a non-deterministic agent doing them by hand. A useful tell: three or more bash commands for one logical operation is a script waiting to be written. And the payoff compounds: the script lands in scripts/ and the next time the job comes up it's already solved. That's often why a good scripts folder accumulates. An agent run leaves nothing reusable behind; a script does.

Spec to the task type. Mechanical, repeatable work wants tight, rigid steps. Judgment work wants the goal and a few guidelines, not a script it'll follow off a cliff. Creative work wants almost nothing but the goal. Over-specify a judgment task and the agent fills your template with "N/A"; under-specify a mechanical one and the output varies wildly run to run.

Protect the main context. Beyond parallelism, the reason to delegate is to keep heavy reads and big tool outputs out of the main thread, the sub-agent absorbs the bulk and hands back a conclusion, not the raw material. Have it log progress to a file as it goes, so a crash or restart doesn't lose the work.

Fit the model where your harness lets you. Cheap model for the grunt extraction, the strong one for the judgment call. This is harness-dependent, some setups (including Claude Code right now) don't reliably route a sub-agent to a different model, so check before you rely on it (see verify-current).

The failure this prevents: spawning an agent for something a script should own (non-determinism where you needed precision, and nothing reusable left behind), mis-specifying it for its task type, or drowning the main thread in a sub-task's output.

---name: agent-delegationdescription: Use when about to delegate a discrete task to a sub-agent, deciding whether to spawn one at all, how much to specify it, and how to keep its work from bloating the main thread. The one-off-task layer below run-a-role-agent. Triggers, "delegate this", "spawn a sub-agent", "should this be a sub-agent or a script", "fan this out".---# Agent-delegation: hand off a task wellBefore you spawn a sub-agent, a couple of decisions decide whether the hand-off helps or just adds cost and flakiness.**First: should this be an agent at all, or a script?** If the work can be described as code, write the code. Agents are for *judgment*; scripts are for *precision*. Copying files, transforming a format, a deterministic sweep, those want a script, not a non-deterministic agent doing them by hand. A useful tell: three or more bash commands for one logical operation is a script waiting to be written. And the payoff compounds: the script lands in `scripts/` and the next time the job comes up it's already solved. That's often *why* a good scripts folder accumulates. An agent run leaves nothing reusable behind; a script does.**Spec to the task type.** Mechanical, repeatable work wants tight, rigid steps. Judgment work wants the goal and a few guidelines, not a script it'll follow off a cliff. Creative work wants almost nothing but the goal. Over-specify a judgment task and the agent fills your template with "N/A"; under-specify a mechanical one and the output varies wildly run to run.**Protect the main context.** Beyond parallelism, the reason to delegate is to keep heavy reads and big tool outputs *out* of the main thread, the sub-agent absorbs the bulk and hands back a conclusion, not the raw material. Have it log progress to a file as it goes, so a crash or restart doesn't lose the work.**Fit the model where your harness lets you.** Cheap model for the grunt extraction, the strong one for the judgment call. This is harness-dependent, some setups (including Claude Code right now) don't reliably route a sub-agent to a different model, so check before you rely on it (see `verify-current`).The failure this prevents: spawning an agent for something a script should own (non-determinism where you needed precision, and nothing reusable left behind), mis-specifying it for its task type, or drowning the main thread in a sub-task's output.

Frequently asked questions