Superpowers
github.com/obra/superpowers- Category
- AI Agents
- Rank
- No. 28Tools index
- Type
- TOOL
- Builder
- obra
- GitHub
- 283.0k stars
- Latest release
- v6.3.0
- Date
About
An agentic skills framework and software-development workflow for coding agents. Built on composable "skills", it makes the agent pause to draw out a spec before writing code, then drives implementation through a structured, repeatable process.
What it does
Superpowers turns a coding agent into a process-driven engineering partner. It asks questions, presents a design for approval, creates a detailed plan, then guides implementation through isolated workspaces, test-first development, review, verification, and branch cleanup. Relevant guidance activates automatically when the host supports startup injection.
Why it's ranked here
The project offers unusually concrete controls against common agent failures: premature coding, vague plans, untested changes, shallow debugging, and unsupported completion claims. Its strongest distinction is end-to-end process coverage across many coding hosts. That rigor also makes it a poor fit for quick edits where ceremony costs more than it saves.
What's good
Plans specify small tasks, exact changes, and verification steps. Implementation can use fresh subagents with separate checks for specification compliance and code quality. The debugging guidance favors root-cause investigation over guessing, while completion requires evidence. Shared, action-based skill text lets host integrations map the same workflow onto their native tools.
Tradeoffs
The workflow is deliberately strict. It mandates test-first implementation, can delete code written before tests, and inserts design approval, planning, review, and branch decisions around coding. Some features degrade when a host lacks subagents or task tracking. Installation is separate for each host, and behavioral evaluations run real agent sessions, take minutes, and are not currently part of continuous integration.
How to use it well
Use it for substantial repository work where requirements need clarification and correctness justifies structured checkpoints. It suits teams that already value tests, isolated branches, review, and explicit verification. Keep a lighter workflow for trivial edits. It does not supply missing host capabilities such as file access, shell execution, or subagents, and it is not an application deployment or project-management service.
Technical notes+
README.md defines the lifecycle from brainstorming through worktree setup, planning, test-driven implementation, review, verification, and branch completion. package.json declares an ECMAScript module package and registers package surfaces for host integration. docs/porting-to-a-new-harness.md describes shared skill bodies, per-host tool mappings, and mandatory automatic session-start injection. docs/testing.md separates plugin infrastructure tests under tests/ from real-session behavioral evaluations under evals/, noting that the latter use Python, tmux, an actor, and a verifier. docs/windows/polyglot-hooks.md documents a Bash-oriented cross-platform hook dispatcher for macOS, Linux, and Windows environments with Git Bash.
Observed
- License
- MIT License
- Package surface
- ECMAScript module package with host-specific plugin and extension registration
- Interfaces
- Coding-agent plugins, extensions, startup hooks, native skill tools, and host tool mappings
- Supported hosts
- Claude Code, Antigravity, Codex App, Codex CLI, Cursor, Devin CLI, Factory Droid, Gemini CLI, GitHub Copilot CLI, Grok Build CLI, Kimi Code, OpenCode, Pi, and Hermes Agent
- Core content
- Harness-agnostic Markdown skills shared across integrations
- Testing structure
- Plugin infrastructure tests and separate real-session skill behavior evaluations
- Platform support
- Hook documentation covers macOS, Linux, and Windows with Git Bash
Read from README.md, package.json, docs/testing.md, docs/README.kimi.md, docs/README.opencode.md, docs/porting-to-a-new-harness.md, docs/windows/polyglot-hooks.md, docs/plans/2026-01-17-visual-brainstorming.md, docs/plans/2025-11-22-opencode-support-design.md, docs/plans/2025-11-22-opencode-support-implementation.md, docs/plans/2025-11-28-skills-improvements-from-user-feedback.md, docs/superpowers/plans/2026-04-06-worktree-rototill.md.
What it can do
Extract project specifications from conversation
User conversation about coding goals → Structured project specification in digestible chunks
Generate implementation plans for software projects
Approved project specification → Step-by-step implementation plan following TDD, YAGNI, and DRY principles
Execute autonomous coding tasks through subagents
Implementation plan and user approval → Working code implementation
Inspect and review code from subagents
Code generated by development subagents → Code quality assessment and corrections
Manage multi-hour autonomous development sessions
Approved project plan → Completed development tasks without user intervention
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.