Guide · Agent skill
How to use the Agent Delegation Prompt Guide agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The agent-delegation skill is a reference document that a coding agent loads before it hands work to a helper: either a subagent with a fresh context or a fork that inherits part or all of the conversation. It shapes how the agent writes that briefing, so the helper gets the goal, the context and the boundaries it needs. This guide explains what the skill changes, how to install it, how to trigger it and where it does not help.
Agent Delegation Prompt GuideQwenLM/qwen-code · Apache-2.0 Download the skill Quick start
- 1Download the ZIP. Get the agent-delegation ZIP from this skill's page on the directory.
- 2Unzip into a skills folder. Extract it into ~/.claude/skills/ for personal use or into the project's .claude/skills/ folder.
- 3Or upload it in the Claude apps. Go to Settings → Capabilities → Skills and upload the skill.
- 4Describe a split task. Ask the agent to divide a larger job between helpers and say which parts need writing and which need research.
- 5Check the briefing. Look for the goal, what is ruled out, boundaries, expected output and whether code should change.
- 6Review the results. Compare what the helpers return with your own understanding before accepting it.
What the skill makes the agent do
Agent skills are folders of instructions that an agent loads when a task calls for them. This one covers only the writing of a delegation prompt. It does not decide whether to delegate, and it does not set the rules for background agents or concurrent runs; those stay in the Agent tool's own description.
The skill pushes the agent toward prompts that explain the goal and why it matters, list what has already been ruled out, give enough context for judgment calls, and state whether the helper should write code or only research. It warns against terse command-style prompts, which the source says produce shallow, generic work.
- Explain what you are trying to accomplish and why.
- Describe what has already been learned or ruled out.
- Give relevant file paths, constraints and what is out of scope.
- Say explicitly whether the helper should write code or do research.
- Ask for a short response when that is what you need.
- Avoid lazy prompts such as 'based on your findings, fix the bug'.
Install the agent-delegation skill
Download the ZIP from the skill's page on this directory and unzip it. The skill folder contains SKILL.md and SKILL.test.ts.
For a personal install, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder. Alternatively, upload the skill in the Claude apps under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Project only: .claude/skills/ inside the project
- Claude apps: Settings → Capabilities → Skills, then upload
How to trigger the skill with natural requests
The skill is written to be loaded before a delegation prompt is written, so you trigger it by asking the agent to hand work to a helper. You do not need a special command. Name the kind of work you want split up and the agent should bring in the reference when it writes the briefing.
Describe the split in plain language. The more concrete the task, the more useful the briefing the agent writes.
- 'Split this refactor across two agents: one maps the call sites, one writes the tests. Brief both properly.'
- 'Fork yourself to investigate why the login tests fail on CI, and tell it what we already ruled out.'
- 'Write a research-only subagent prompt to find every place we read the config file.'
A worked example of a good versus lazy briefing
The source's worked example shows an agent writing a prime-checking function and then launching a test-runner subagent after the code is done. The point for delegation is the same: the helper needs to know what was written, where, and what to check.
A lazy briefing says 'based on your findings, fix the bug'. A better one names the file, the failing behaviour, what was already tested, the constraint to keep the public API unchanged, and says the helper should write the fix and report the diff summary.
- Lazy: 'Based on your findings, fix the bug.'
- Better: names the file path, the symptom, what was ruled out, the allowed change and what is out of scope.
Know what the skill cannot override
If you delegate to a custom subagent that someone has defined, that definition already sets its scope, tools and output format. Your prompt adds task context but cannot widen those limits. For example, a subagent confined to one file cannot search the repository just because the prompt asks it to.
If the subagent's contract is too narrow for your job, change its definition rather than the prompt.
Safety notes: scripts and network
According to the listing, this skill runs no scripts and uses no network access. It needs nothing else to be installed. It is a text reference that changes how the agent writes prompts, so the safety question is mostly about what the helpers you launch are allowed to do, which is set by the Agent tool and any custom subagent definitions you use.
- Runs scripts: no
- Uses network: no
- Requires: nothing listed
Tips for better delegated work
Use delegation for larger tasks where separate agents can work in parallel or where a fresh context helps. For a simple single-step job, the overhead of writing a full briefing is not worth it.
Be explicit about scope in fork prompts. With full history the prompt should be a directive; when history is limited, include the older context the fork still needs.
- State the output you expect and its length.
- Name what another agent is already handling so work does not overlap.
- Check returned results against your own understanding before acting on them.
What people use it for
- Splitting a multi-file refactor so one agent maps call sites while another writes tests.
- Forking an agent to investigate a failing CI test while the main session continues.
- Launching a research-only subagent to find every place a config file is read.
- Writing consistent briefings for a team's custom subagent definitions.
Frequently asked questions
How do I install the agent-delegation skill?
Download the ZIP from this skill's page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. You can also upload it in the Claude apps under Settings → Capabilities → Skills.
Does the agent-delegation skill require anything else?
The listing says it requires nothing else, runs no scripts and uses no network access. It is a reference document that the agent reads.
Is the agent-delegation skill free?
The skill is listed under the Apache-2.0 licence. Check the licence file in the source repository for the terms that apply to your use.
Does this skill make the agent delegate tasks automatically?
No. It only changes how the agent writes a prompt once it has decided to delegate. Whether to delegate and the rules for background agents are set in the Agent tool's own description.
Why does my subagent ignore parts of my prompt?
If you are using a custom subagent, its own definition fixes its scope, tools and output contract. A prompt cannot widen those limits, so the definition has to change if the contract is too narrow.
How is this different from other prompt guides?
This skill is narrow: it covers only the briefing for subagents and forks. It does nothing on its own and is not a standalone tool for non-developers.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you run several agents on a larger coding task or maintain custom subagent definitions and want more consistent, less shallow delegated work. Skip it for simple single-step tasks, where delegating adds overhead, or if you want a standalone tool, since it only changes how the agent writes prompts.
Download the skill