Superpowers: A Full Software Methodology Packaged as Agent Skills

What Superpowers Actually Is

Superpowers is not a library or a CLI tool. It is a set of instructions and composable skills that reshape how a coding agent behaves. The README calls it 'a complete software development methodology for your coding agents.' The core idea is that the agent does not jump into code. Instead, it steps back, asks what you are trying to do, writes a spec, shows it to you in digestible chunks, then produces an implementation plan. Only after you say 'go' does it start writing code, and even then it uses a subagent-driven process where agents work through tasks, inspect their work, and review it. The target user is a developer who wants their agent to follow a repeatable engineering practice rather than improvise. The primary language is Shell, which suggests the skills are largely script-based hooks and prompts, not a compiled runtime.

The Workflow: From Spec to Subagents

The mechanism is a staged workflow that triggers automatically. According to the README, the agent starts by asking what you are really trying to do, then teases out a spec and shows it to you 'in chunks short enough to actually read and digest.' After you sign off, the agent builds an implementation plan 'clear enough for an enthusiastic junior engineer with poor taste, no judgement, no project context, and an aversion to testing to follow.' That plan emphasizes red/green TDD, YAGNI, and DRY. Once you say 'go,' the agent launches a 'subagent-driven-development' process. Subagents work through each engineering task, inspect and review their work, and continue. The README claims it is common for the agent to work autonomously for a couple of hours without deviating from the plan. The skills trigger automatically, so you do not need to invoke them manually. This is a significant departure from typical agent usage, where you prompt for each step.

Installation Across Harnesses

Installation differs by harness, and the README is explicit that you must install Superpowers separately for each one. For Claude Code, you can install from Anthropic's official marketplace with '/plugin install superpowers@claude-plugins-official' or from the Superpowers marketplace. For Antigravity, the command is 'agy plugin install https://github.com/obra/superpowers'. Codex CLI uses '/plugins' and searches for 'superpowers'. Cursor uses '/add-plugin superpowers' in chat. Devin CLI uses 'devin plugins install obra/superpowers'. Gemini CLI uses 'gemini extensions install https://github.com/obra/superpowers'. The list also includes GitHub Copilot CLI, Grok Build CLI, Kimi Code, OpenCode, Pi, and Hermes Agent. The variety is a strength for multi-tool users, but it also means you have to learn a different install command for each environment. The README warns that OpenCode uses its own plugin install, so even if you use Superpowers in another harness, you must install it separately.

The Bootstrap and Hook Dependency

A key mechanism is the bootstrap that injects the 'using-superpowers' skill at session start. For Pi, the README says the package loads the skills and 'a small extension that injects the using-superpowers bootstrap at session startup and again after compaction.' For Hermes, the README notes a limitation: 'Hermes has no post-compaction hook, so a very long session that compacts over its first turn' will lose the bootstrap. The text is truncated, but the implication is that you must manually re-inject the bootstrap after compaction. This is a concrete failure mode. If you use a harness without a post-compaction hook, a long session can silently lose the methodology, and the agent may revert to its default behavior. That is a real operational cost, and it is not obvious from the marketing language. You need to know your harness's hook support before relying on Superpowers for extended autonomous sessions.

What's Inside the Skills

The README's table of contents lists the covered areas: design, planning, testing, debugging, review, and branch completion. The philosophy section is truncated, but the workflow description gives a clear picture. The skills are composable, meaning you can mix and match them, but the default workflow chains them in a specific order. The emphasis on TDD, YAGNI, and DRY is a deliberate opinion. That opinion is not neutral. If you prefer a more exploratory coding style or work in a codebase where strict TDD is impractical, Superpowers will fight you. The README also mentions a 'Visual companion telemetry' section, which suggests there is some kind of telemetry or visual feedback, but the README does not explain what it collects or how. That is a gap you should investigate before using it in a sensitive environment. The MIT license is permissive, so you can modify the skills, but the README does not detail how to customize them.

Maintenance and Upgrade Cost

The repository has recent releases, with v6.3.0 pushed on 2026-08-12, and the README includes an 'Updating' section, though its content is truncated. For Devin CLI, the README gives an explicit update command: 'devin plugins update superpowers'. For Gemini CLI, it is 'gemini extensions update superpowers'. For Antigravity, the README says 'Reinstall with the same command to update.' This means updates are manual per harness. There is no central updater. If you use Superpowers across five harnesses, you have to remember to update each one separately. The release cadence appears active, with three versions in roughly six weeks, so updates are frequent. That is good for fixes but adds maintenance overhead. The license is MIT, which is permissive for commercial use, but the README also mentions commercial services from Primeradiant for enterprise support, additional tooling, or managed spending. That is a separate offering, not part of the open-source project.

Alternatives and When to Choose Otherwise

The most direct alternative is not a single tool but the default behavior of your coding agent without any methodology layer. Most agents, like Claude Code or Codex, will happily write code on request. Superpowers exists to override that default. Another alternative is a custom set of instructions or a system prompt that you write yourself. That approach gives you full control but lacks the composable skills and the subagent orchestration. A third alternative is a lighter plugin that only adds a few prompts, such as a code review skill, without enforcing a full spec-first workflow. The difference is in the degree of enforcement. Superpowers is a full methodology that changes the agent's entire interaction pattern. A lightweight prompt does not. If you only need code review, installing Superpowers is overkill. If you want your agent to run autonomously for hours, the subagent-driven approach is the core differentiator, but it also depends on the subagent and task-list tools being available in your harness, which the README notes are 'optional Pi companion packages' for Pi.

Editorial conclusion

Adopt Superpowers if you want your coding agent to follow a strict spec-first, TDD-driven workflow and you are comfortable with a methodology that overrides the agent's default behavior. Skip it if you need a lightweight skill pack or if your team already has a strong process that conflicts with its opinionated phases. Before adopting, verify that your harness supports session-start and post-compaction hooks, because the README notes that some harnesses, like Hermes, lack these and require manual re-injection. Also confirm that the plugin marketplace for your tool is actively maintained, since installation paths differ per harness and updates are manual per the README.

Official sources

1. Official README
2. Project repository
3. Release notes