Guide · Agent skill
How to use the Dispatching Parallel Agents agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Dispatching Parallel Agents skill teaches an AI agent to split several unrelated problems, such as failing tests in different files, across separate sub-agents that work at the same time. It is for developers with independent bugs who want them handled faster, and for anyone learning how to write focused instructions for delegated agents. After this guide you will know how to install the skill, how to ask for it, and when to skip it.
Dispatching Parallel Agentsobra/superpowers · MIT Download the skill Quick start
- 1Download the ZIP. Download the skill package from this page and unzip it. It contains SKILL.md.
- 2Place the skill folder. Move the folder to ~/.claude/skills/ for all projects, or to a project's .claude/skills/ folder for one project.
- 3Or upload it in the apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
- 4List your independent problems. Write down each failing test file or bug and what part of the code it touches.
- 5Ask for parallel work. Describe the problems and ask the agent to split them across sub-agents, one per independent domain.
- 6Check the results. Read each summary, look for overlapping edits, and run the full test suite before accepting the changes.
What the skill makes the agent do
When an agent faces two or more independent tasks, the skill has it hand each one to its own sub-agent. Each sub-agent gets its own isolated context and a carefully written instruction, so it does not inherit the main session's history. The main agent keeps its own context free for coordinating and checking results.
The skill's core principle is to dispatch one agent per independent problem domain and let them work concurrently. Its example is three test files failing for different reasons, such as tool approval, batch completion and abort handling.
Install the skill from our download page
Download the skill as a ZIP from this page and unzip it. The package contains a single file, SKILL.md. Place the unzipped folder in one of two locations:
Use ~/.claude/skills/ to make the skill available across your projects, or use a project's .claude/skills/ folder to limit it to that project.
You can also upload the skill in the Claude apps under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/dispatching-parallel-agents
- Project: .claude/skills/dispatching-parallel-agents
Trigger the skill with natural requests
The skill's description tells the agent to use it when facing two or more independent tasks that can be worked on without shared state or sequential dependencies. You do not need a special command. Describe the situation plainly and say you want the work split up or done in parallel.
Name the problems clearly. The skill works best when you list each failure or bug and where it lives.
- Several test files are failing in different modules. Can you split these across agents and fix them in parallel?
- There are three unrelated bugs: login redirect, invoice export and search indexing. Handle each one with its own sub-agent.
- Fix these failing tests, one agent per file, and tell me what each one changed.
Write good sub-agent instructions
The skill's pattern gives each agent a focused scope, a clear goal, explicit constraints and a defined output. Each instruction must be self-contained, so the agent understands the problem without seeing the rest of your session.
The skill also lists common mistakes. Asking to fix all the tests is too broad. Saying fix the race condition gives no location. Leaving out constraints can lead an agent to refactor everything. Asking for a vague result leaves you unsure what changed.
- Scope: one test file or subsystem
- Goal: make these tests pass
- Constraints: for example, do not change production code, or fix tests only
- Output: a summary of root cause and changes
- Paste the exact error messages and test names into the instruction
Fix the 3 failing tests in src/agents/agent-tool-abort.test.ts:
1. "should abort tool with partial output capture" - expects 'interrupted at' in message
2. "should handle mixed completed and aborted tools" - fast tool aborted instead of completed
3. "should properly track pendingToolCount" - expects 3 results but gets 0
Do NOT just increase timeouts - find the real issue.
Return: Summary of what you found and what you fixed.Review and check the results
When the agents return, read each summary, check whether any of them edited the same code, run the full test suite and integrate the changes. The skill also notes that agents can make systematic errors, so spot-checking matters.
- Review each summary to understand what changed
- Check for conflicts, such as agents editing the same code
- Run the full suite to verify all fixes work together
- Spot-check results, since agents can make systematic errors
Know when not to use it
Do not use the skill when failures are related, because fixing one may fix the others. Investigate them together first. Also avoid it when you need to understand the whole system, when you do not yet know what is broken, or when agents would edit the same files or use the same resources.
- Related failures with a shared root cause
- Work that needs full-system context
- Exploratory debugging where the problems are not yet identified
- Tasks that edit the same files or share resources
What people use it for
- Fixing failing tests spread across several unrelated test files after a refactor
- Handling separate bugs in different subsystems, such as login, billing and search, at the same time
- Learning how to write scoped, self-contained instructions for delegated AI agents
- Deciding whether a set of failures is independent enough to split or should be investigated together
Frequently asked questions
How do I install the dispatching-parallel-agents skill?
Download the ZIP from this page and unzip it. Place the folder in ~/.claude/skills/ for personal use or in a project's .claude/skills/ folder for one project. You can also upload it in the Claude apps under Settings → Capabilities → Skills.
Does this skill need anything else installed?
The source lists no requirements. It runs no scripts and uses no network access. It is a single SKILL.md file that gives the agent instructions.
Is the dispatching-parallel-agents skill free?
The skill is published under the MIT licence. This page does not list prices, so check the directory listing for any other terms.
When should I not use parallel agents?
Avoid them when failures share a root cause, when tasks edit the same files or share resources, or when you do not yet know what is broken. The skill itself warns to investigate related failures together first.
Why did my agents conflict or fix the same thing twice?
This usually means the problems were not truly independent. Check whether the failures share a cause and whether agents edited the same files. Next time, group related failures before dispatching.
How is this different from asking one agent to fix everything?
A single agent works through problems in sequence and uses one context for all of them. This skill has each problem handled by a separate sub-agent with its own focused instructions, so independent work can happen at the same time and the main agent stays free to coordinate.
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 have several unrelated failures or bugs in different modules and an agent setup that supports sub-agents. It is also useful for learning how to scope instructions tightly. Pick something else for debugging where failures share a cause, where tasks touch the same files, or where you have not yet identified the problems.
Download the skill