Dispatching Parallel Agents Skill

Dispatching Parallel Agents is a skill from Superpowers (skill source). It applies when an agent faces two or more independent tasks with no shared state or ordering, such as several unrelated test failures, and shows how to hand each one to its own subagent at the same time instead of investigating them one after another.

Core principle: one agent per independent problem domain, working concurrently.

Key Features

• Decision flow: are there multiple problems, are they independent, and can they run without shared state? Related failures go to a single agent; shared state means sequential agents.
• Focused tasks: each subagent gets a specific scope (one test file or subsystem), a clear goal, constraints such as "do not change other code", and the expected output.
• Crafted context: subagents never inherit the session's history; the coordinator writes exactly what each one needs, which also preserves its own context.
• Parallel dispatch: all subagent calls go in the same response; one call per response runs them sequentially.
• Integration: read every summary, check that fixes do not conflict, then run the full test suite.
• Worked example: six failures across three test files after a refactor, fixed by three agents with no conflicts.

Use Cases

• Several failing test files with different root causes.
• Independent subsystems broken at once.
• Research questions that do not depend on each other.

Pricing

Free and open source under the MIT license. Prime Radiant, the company behind Superpowers, sells commercial support to enterprises. The practical cost is tokens: process skills add questions, reviews and subagent runs.

Getting Started

Superpowers installs as one plugin, so this skill arrives with the rest of the library (15 skills in v6.4.1). In Claude Code run /plugin install superpowers@claude-plugins-official. Cursor uses /add-plugin superpowers, Gemini CLI uses gemini extensions install https://github.com/obra/superpowers, and the README lists commands for Codex, GitHub Copilot CLI, OpenCode, Pi and other harnesses. Skills trigger on their own; you can also ask for one by name. Describe several independent problems and ask the agent to handle them in parallel.

Limitation: it needs a harness that runs subagents concurrently, and parallel agents multiply token spend. The skill warns against using it when failures are related or when agents would edit the same files.

FAQ

Is this the same as Subagent-Driven Development?

No. That skill executes a plan task by task with reviews; this one fans out independent problems at once.

How do I avoid conflicting edits?

Give each agent a disjoint scope, or isolate them with worktrees.

Alternatives

• Claude Subagents: Claude Code's native specialist workers.
• Agent Dispatcher: routes tasks to specialist roles.
• Conductor: runs parallel Claude Code instances in a desktop app.

Conclusion

A clear recipe for parallelism that stays safe because it insists on independence first. More in the skills hub.