What Is the Superpowers Plugin? Understanding OpenAI's Agent Workflow Framework

The superpowers plugin is a comprehensive software-development methodology packaged as a Codex-style plugin that supplies coding agents with a library of composable skills and a bootstrapped workflow driving every phase from idea generation to final branch cleanup.

The superpowers plugin, hosted in the openai/plugins repository, transforms generic coding agents into disciplined software-engineering partners by codifying the entire development lifecycle into reusable, searchable behaviors. Unlike ad-hoc agent instructions, this framework ensures consistent execution of critical phases—from brainstorming to test-driven development—across Claude Code, Cursor, Devin, Gemini, and twelve other AI coding platforms.

Core Architecture of the Superpowers Plugin

The superpowers plugin operates through three tightly integrated components that enforce methodological rigor without bloating token usage.

The Skills Library and Composable Behaviors

At the heart of the system lies a markdown-based skills library located in plugins/superpowers/skills/. Each skill definition—exemplified by plugins/superpowers/skills/writing-skills/SKILL.md—contains:

• YAML front-matter with name and description fields that function as contextual triggers
• Trigger descriptions that must start with "Use when..." so agents can search and match skills to specific development contexts
• Concrete patterns and examples that remain on disk until explicitly loaded

This architecture delivers token efficiency by injecting only the trigger description into the context window by default; the heavy reference material stays on disk and is fetched on demand when an agent selects a particular skill.

Bootstrap Hook and Mandatory Activation

The plugin employs a bootstrap extension named using-superpowers that injects skill-loading logic at two critical moments:

1. Session initialization—guaranteeing the workflow is active from the first prompt
2. After compaction events—ensuring skills persist across context window resets

This mandatory activation eliminates "ad-hoc" behavior where agents might skip critical steps like code review or branch cleanup. Agents never need to remember to manually import the plugin or invoke the workflow.

Telemetry and Enterprise Layer

The implementation in plugins/superpowers/README.md (lines 44-48) includes an optional telemetry system that reports only the plugin version—never project data—enabling maintainers to track adoption while respecting privacy. Users can disable this via:

export SUPERPOWERS_DISABLE_TELEMETRY=1

The repository also references a commercial service layer for enterprises requiring SLA-backed assistance and managed deployment options.

Why the Superpowers Plugin Matters

The significance of this framework extends beyond mere task automation, addressing fundamental reliability issues in AI-assisted development.

Consistency Across Agents. By codifying the entire development lifecycle—from brainstorming → worktree creation → writing-plans → sub-agent-driven development → test-driven development → code-review → branch finalisation—the plugin eliminates variability that leads agents to skip critical verification steps.

Token Efficiency. The selective loading mechanism ensures that only relevant skill triggers occupy the context window, while detailed reference material remains disk-resident until needed.

Cross-Platform Reach. The same plugin installs across Antigravity, Claude Code, Codex App/CLI, Cursor, Devin, Gemini, GitHub Copilot CLI, Grok Build, Kimi Code, OpenCode, Pi, and Hermes Agent through unified installation commands documented in the README.

Extensibility. New skills can be added to plugins/superpowers/skills/ without modifying core plugin logic, allowing ecosystem evolution while preserving backward compatibility.

Robust Testing via RED-GREEN-REFACTOR. Every skill in the library follows strict test-driven development cycles, ensuring agents have been validated under pressure scenarios before deployment.

Installing and Using the Superpowers Plugin

The plugin supports multiple installation vectors depending on your agent harness.

Cross-Platform Installation Examples

For Antigravity, install via:

agy plugin install https://github.com/obra/superpowers

Re-run the same command to update the plugin to the latest version.

For Claude Code, use the plugin registry:

/plugin install superpowers@claude-plugins-official

Once installed, the agent automatically runs the brainstorming skill when you initiate a conversation about a new codebase.

Manual Skill Triggering

While the bootstrap hook handles automatic activation, you can manually invoke specific skills:

/use-superpowers brainstorming

This command triggers a structured workflow where the agent:

1. Asks clarifying questions about the problem domain
2. Presents a short design sketch
3. Waits for explicit sign-off before proceeding to the next workflow stage

The underlying implementation resides in the brainstorming skill definition within the Skills Library.

Key Implementation Files

Understanding the superpowers plugin requires familiarity with these specific source files:

• plugins/superpowers/README.md—Contains the full workflow documentation, installation matrix for all supported platforms, and the "basic workflow" that agents execute automatically
• plugins/superpowers/skills/writing-skills/SKILL.md—Illustrates the canonical skill definition format, including YAML front-matter structure and TDD-style process requirements for skill creation
• plugins/superpowers/.opencode/INSTALL.md—Provides the installation script specific to the OpenCode harness
• plugins/superpowers/LICENSE—MIT license governing reuse and distribution
• plugins/superpowers/assets/superpowers-small.svg—Icon asset used in UI integrations for Claude, Cursor, and other visual interfaces

Summary

• The superpowers plugin is a Codex-style methodology framework that converts generic agents into disciplined software engineers through composable skills and mandatory workflows.
• Skills Library in plugins/superpowers/skills/ uses "Use when..." trigger descriptions to enable efficient, searchable behavior loading without token bloat.
• Bootstrap hook using-superpowers ensures mandatory activation at session start and after compaction, preventing agents from skipping critical phases.
• The plugin supports twelve+ platforms including Claude Code, Cursor, and Devin via unified installation commands.
• Optional telemetry (disable via SUPERPOWERS_DISABLE_TELEMETRY) reports version-only data for adoption metrics while preserving project privacy.
• All skills follow RED-GREEN-REFACTOR test-driven development cycles to ensure reliability under pressure.

Frequently Asked Questions

How does the superpowers plugin differ from standard agent system prompts?

Unlike static system prompts that provide general instructions, the superpowers plugin implements a bootstrapped workflow through the using-superpowers hook that dynamically loads specific skills based on context. This ensures agents follow the complete development lifecycle—from brainstorming to branch cleanup—rather than improvising steps ad-hoc. The plugin also enforces token efficiency by loading only trigger descriptions initially, whereas standard prompts often include entire rulebooks in every context window.

What AI coding platforms support the superpowers plugin?

According to the installation matrix in plugins/superpowers/README.md, the plugin supports Claude Code, Antigravity, Codex App/CLI, Cursor, Devin, Gemini, GitHub Copilot CLI, Grok Build, Kimi Code, OpenCode, Pi, and Hermes Agent. Each platform uses platform-specific installation commands (e.g., agy plugin install for Antigravity, /plugin install for Claude), but the underlying skill library and workflow remain identical across all implementations.

How do I disable telemetry in the superpowers plugin?

Set the environment variable SUPERPOWERS_DISABLE_TELEMETRY to any value before starting your agent session. As documented in lines 44-47 of the README, the telemetry system reports only the plugin version number for adoption metrics and never transmits project data, code, or user prompts. Disabling this variable prevents even the version ping from being sent.

Can I extend the superpowers plugin with custom skills?

Yes. The architecture in plugins/superpowers/skills/writing-skills/SKILL.md demonstrates that new skills require only a markdown file with YAML front-matter containing name and description fields (where the description starts with "Use when..."), followed by the skill's overview and concrete examples. Adding files to the skills directory extends the plugin without modifying core logic, maintaining backward compatibility while allowing organization-specific methodologies to integrate with the existing RED-GREEN-REFACTOR workflow.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}