This is a feature request for 'obra/superpowers', a tool for AI-driven development. The request is to add support for Claude Code's experimental Agent Teams feature, which allows multiple AI agents to coordinate with shared task lists and inter-agent messaging. The current 'superpowers' skills are not aware of these new primitives, leading to conflicts and a lack of orchestration capabilities for team leads.
A starter prompt for Claude Code, what you'll need, and how to reach them.
You are an expert software engineer specializing in AI agent development and tool integration. Your task is to implement support for Claude Code Agent Teams within the `obra/superpowers` framework. The goal is to enable `superpowers` skills to recognize and utilize the `Teammate`, `SendMessage`, `TaskCreate`, `TaskUpdate`, and `TaskList` tools, and to create a new orchestration skill for team leads. **Framework:** Fork `https://github.com/obra/superpowers` (the `main` branch). The existing project uses Python. Maintain code consistency with the existing codebase. **Core Requirements:** 1. **Tool Recognition:** Modify existing `superpowers` skills (e.g., `executing-plans`, `subagent-driven-development`) to correctly interpret and use `Teammate`, `SendMessage`, `TaskCreate`, `TaskUpdate`, and `TaskList` when interacting with Claude Code Agent Teams. 2. **Agent Team Orchestration Skill:** Implement a new, dedicated `superpowers` skill named `orchestrating-agent-teams`. This skill should guide the Claude Code team lead agent on: - How to break down complex problems into sub-tasks suitable for team execution. - How to dynamically spawn and manage teammates using the `Teammate` tool. - How to assign tasks to specific teammates using team-aware `TaskCreate`. - How to facilitate inter-agent communication via `SendMessage`. - How to handle plan approvals and coordination between team members. - How to coordinate the overall shutdown of the agent team once tasks are complete. 3. **Conflict Resolution:** Ensure the new `orchestrating-agent-teams` skill coexists without conflict with existing parallel-work skills like `dispatching-parallel-agents`. **MVP Slice:** Focus first on implementing basic recognition of the new Agent Team tools within `executing-plans` and building a minimal `orchestrating-agent-teams` skill that can successfully spawn two teammates, assign them distinct simple tasks (e.g., 'write a short python function for X'), receive messages, and signal completion. **Build/Verify Gate:** The solution is complete when a series of unit and integration tests (using `pytest` or `unittest` as appropriate for the `superpowers` project) pass, demonstrating that a Claude Code team lead, using the new `superpowers` skills, can successfully coordinate a two-agent team to complete a simple, decomposed task, and that existing `subagent-driven-development` functionality remains intact. **Deliverable:** A pull request to the `obra/superpowers` repository with the new functionality, tests, and documentation updates.
Reach out to the `obra/superpowers` community and AI developers using Claude Code to highlight how your `mcp-kit` or `forge-kit` can help them build and manage similar advanced AI agent tooling, and offer your contribution to `superpowers` as proof of expertise.
## Is your feature request related to a problem? Please describe. Claude Code 2.1.32+ introduced experimental **Agent Teams** — a feature that allows coordinating multiple Claude Code instances as a team with a lead, teammates, shared task lists, and inter-agent messaging. See the [official documentation](https://code.claude.com/docs/en/agent-teams). Superpowers currently has skills for subagent-driven development (`subagent-driven-development`, `dispatching-parallel-agents`) but no awareness of or support for the newer Agent Teams primitives. When using superpowers with agent teams, several issues arise: 1. **Skills don't recognize team context.** Skills like `executing-plans` and `subagent-driven-development` only know about the `Task` tool (subagents), not `Teammate`, `SendMessage`, or the team-aware `TaskCreate`/`TaskUpdate`/`TaskList` tools. 2. **No skill for orchestrating agent teams.** There's no skill that guides the team lead on how to break down work, spawn teammates, assign tasks, handle plan approvals, and coordinate shutdown — the way `dispatching-parallel-agents` does for subagents. 3. **Existing parallel-work skills may conflict.** `dispatching-parallel-agents` spawn
Standard for any OSS contribution
Operator is familiar with Claude products.
Standard for `superpowers` project.
New experimental feature; requires reviewing official documentation - ~1-2 hours.
Learn it: Search getting-started ↗
Get set up: Create the account/instance, generate the API key or credentials, and add them to your project's environment variables.
Requires diving into existing `superpowers` skill implementation - ~0.5-1 day.
Learn it: Search getting-started ↗
Comment on the GitHub issue #429 in the `obra/superpowers` repository.
“I've reviewed the Agent Teams feature request for `obra/superpowers` and have a working plan to implement it. I'm an experienced AI agent developer and can deliver a PR that integrates `Teammate`, `SendMessage`, and `TaskList` tools, along with a new orchestration skill. I'll open a draft PR soon to demonstrate progress.”
Open the original ↗