RepoDaily · 2026-07-01 · Infrastructure / Runtime

Superpowers: A Skills Framework That Turns Coding Agents Into Disciplined Engineers

#6 Infrastructure / Runtime Shell +884 obra/superpowers Open repository

Superpowers (v6.1.0, MIT) installs a spec-first, TDD-enforcing development methodology into Claude Code, Cursor, Codex, and seven other agent harnesses through per-platform plugins.

Repo typeInfrastructure / Runtime
Best forDevelopers who use AI coding agents (Claude Code, Cursor, Codex, etc.) and want the agent to plan, spec, and test before writing code rather than rushing to generate files.
Risk levelMedium — behavioral methodology that changes how agents operate in your repo; requires trust and review of auto-triggered skills.
Time to evaluate30–60 minutes to install on one harness and run a small feature through the spec → plan → implement loop.

Primary question: Are you comfortable letting an agent spend time on specs and subagent-driven planning before producing code?

90/100

RepoDaily adoption score

RepoDaily rates this as 90/100 (strong) for adoption: evidence, installation path, production risk, differentiation, license clarity, and AI/agent fit are scored from the article sources and adoption notes.

Directional score from RepoDaily sources and adoption notes, not a benchmark.Risk: Medium
100Evidence quality

4 source(s) across 4 source category/categories, plus a RepoDaily-specific evidence module when available.

93Installability

7 workflow step(s), 5 next-action step(s), and 0 command/install signal(s) were detected.

67Maintenance confidence

Trending momentum is +884 stars, with maintenance/release/issue signals counted when present.

90Production readiness

Risk is marked medium, with 4 security note(s) and 4 explicit skip condition(s).

100Differentiation

3 opportunity lens item(s), 4 alternative(s), and 4 type-specific section(s) support differentiation.

82License clarity

License source or license wording is present.

90Agent / AI fit

7 AI/agent-related signal(s) were detected in the article text and metadata.

Project overview

Superpowers is a Shell-based agentic skills framework and software development methodology created by Jesse Vincent. Version 6.1.0 ships as an MIT-licensed package that you install into an existing coding-agent harness — it is not a standalone model or editor. Once installed, it intercepts the agent's workflow so that, instead of immediately writing code, the agent first asks what you are trying to accomplish, drafts a spec in reviewable chunks, builds an implementation plan, and only then begins coding through a subagent-driven process.

The project distinguishes itself from prompt snippets and system-prompt tweaks by packaging composable skills plus session-start hooks. The skills trigger automatically; there is no manual slash-command ritual the developer must remember each turn. The README states agents can work autonomously for a couple of hours at a time without deviating from the agreed plan.

Support is broad: the Quickstart lists Claude Code, Antigravity, Codex App, Codex CLI, Cursor, Factory Droid, GitHub Copilot CLI, Kimi Code, OpenCode, and Pi. Installation differs per harness and must be repeated for each one you use. The package.json confirms the project is an ES module (`"type": "module"`) with its main entry at `.opencode/plugins/superpowers.js`, and declares a Pi extension at `./.pi/extensions/superpowers.ts` plus a skills directory at `./skills`.

Commercial backing exists: PrimeRadiant offers enterprise support and managed spending (sales@primeradiant.com), and the project is actively hiring a full-time community engineer, signaling ongoing maintenance intent rather than a weekend experiment.

Problem it solves

  • AI coding agents frequently jump straight to writing code without confirming intent, leading to wasted output and misaligned implementations.
  • Without an enforced methodology, agents skip tests, ignore DRY/YAGNI principles, and produce plans too vague for reliable execution.
  • Switching harnesses (Claude Code to Cursor, Codex to OpenCode) usually means losing your custom prompts and behavior each time.
  • Enterprise teams adopting AI agents lack a consistent, reviewable development process that junior and senior engineers can share.
  • Manual prompt engineering does not scale across team members because each person reinvents the spec-and-plan workflow.

How it works

  1. Install Superpowers into your chosen agent harness using its platform-specific command (e.g., `/plugin install superpowers@claude-plugins-official` for Claude Code or `/add-plugin superpowers` for Cursor).
  2. When you start a session and describe a build task, the session-start hook activates the skills automatically — no manual trigger is required.
  3. The agent steps back and asks clarifying questions to tease out a spec from the conversation rather than immediately generating code.
  4. The spec is presented in short, reviewable chunks; you sign off on the design before any implementation begins.
  5. The agent produces a detailed implementation plan that the README describes as clear enough for an enthusiastic junior engineer with poor taste and no project context to follow.
  6. After you say 'go', a subagent-driven-development process launches: subagents work through each engineering task while the orchestrating agent inspects and reviews their output.
  7. The loop emphasizes true red/green TDD, YAGNI, and DRY, and can run autonomously for hours as long as tasks stay within the approved plan.

Architecture Read: Plugin Bootstrap and Skills Layout

package.json declares Superpowers as an ES module (`"type": "module"`) named `superpowers` at version `6.1.0`, with `main` set to `.opencode/plugins/superpowers.js`. The Pi integration block references `./.pi/extensions/superpowers.ts` for the extension layer and `./skills` as the skills directory, confirming that composable skills are file-system-backed rather than hard-coded into a single script.

For harnesses like Antigravity, the README explains that the plugin's session-start hook runs automatically, which is why the skills are active from the first message. Reinstalling with the same `agy plugin install` command updates the plugin in place. This hook-based design is what allows the methodology to feel invisible to the end user.

The package keywords — `pi-package, skills, tdd, debugging, collaboration, workflow` — signal that the framework treats TDD, debugging, and collaboration as first-class skill categories rather than afterthoughts bolted onto a code generator.

Command Surface Across Harnesses

  • Claude Code (official Anthropic marketplace): `/plugin install superpowers@claude-plugins-official`
  • Claude Code (Superpowers marketplace): `/plugin marketplace add obra/superpowers-marketplace`, then `/plugin install superpowers@superpowers-marketplace`
  • Antigravity: `agy plugin install https://github.com/obra/superpowers` (session-start hook auto-activates)
  • Codex CLI: open `/plugins`, search `superpowers`, select `Install Plugin`
  • Cursor: `/add-plugin superpowers` in Agent chat, or search 'superpowers' in the plugin marketplace
  • Codex App: open Plugins sidebar, find Superpowers in the Coding section, click `+`

Try-It Path: 30-Minute Evaluation

Pick a single harness you already use daily — Claude Code is the lowest-friction option because Superpowers is listed in Anthropic's official marketplace. Run `/plugin install superpowers@claude-plugins-official` and restart the session.

Describe a small, well-scoped feature (for example, a single utility function with clear inputs and outputs). Observe whether the agent pauses to ask spec questions and returns chunks for review rather than dumping code.

Approve the spec, approve the plan, then say 'go'. Watch the subagent loop execute at least one red/green TDD cycle. If the agent produces a failing test before the implementation, the methodology is installed and functioning correctly.

Maintenance Risk and Commercial Backing

The project is actively backed by PrimeRadiant, which offers enterprise support and managed spending via sales@primeradiant.com. The README also announces a full-time community-engineer hiring round (primeradiant.com/jobs/superpowers-community-engineer/), which reduces the risk of the project being abandoned.

Version 6.1.0 with ten supported harnesses implies ongoing maintenance load: each harness plugin surface can change independently. Teams adopting Superpowers should track per-harness changelog entries before upgrading.

License is MIT (Copyright 2025 Jesse Vincent), so there are no commercial-use restrictions if internal fork-back becomes necessary.

Who should pay attention?

Good fit if

  • Solo developers and small teams using Claude Code, Cursor, or Codex who want enforced spec-first behavior
  • Engineering leads standardizing how AI agents operate across a team (same skills, same TDD discipline)
  • Developers frustrated by agents that skip tests and produce unreviewed plans
  • Teams already committed to red/green TDD, YAGNI, and DRY who want their agent to follow the same rules
  • Organizations that can benefit from PrimeRadiant's commercial support layer

Skip for now if

  • Developers who do not use any of the ten supported agent harnesses
  • Teams with strict policies against session-start hooks or auto-triggered agent skills
  • Projects where the agent must produce code immediately with no spec or planning step
  • Users who need a standalone editor or model rather than a plugin layer

Risks and cautions

Medium

The framework changes how your agent behaves inside your repository by injecting spec, plan, and TDD steps before code generation. The risk is behavioral and trust-related, not licensing.

  • Session-start hooks auto-activate skills, which can surprise teams unfamiliar with the methodology
  • Subagent-driven development can run autonomously for hours, requiring confidence in the plan and monitoring of token spend
  • Ten harness surfaces means ten potential upgrade paths; a harness plugin API change can break one surface independently
  • Enterprise teams must evaluate whether commercial support from PrimeRadiant is needed before broad rollout
  • License is MIT (Copyright 2025 Jesse Vincent) with no additional clauses restricting commercial use or modification.
  • Plugin installation pulls from obra/superpowers or the obra/superpowers-marketplace GitHub repositories; teams should pin and audit the installed commit.
  • Session-start hooks execute on every session, so reviewers should inspect the hook content before deploying across an organization.
  • package.json entry point is `.opencode/plugins/superpowers.js` and the Pi extension is `./.pi/extensions/superpowers.ts`; both are file-system readable for security review.

Alternatives to compare

ApproachWhen to useTrade-off
Aider
You want a standalone terminal coding agent with built-in git workflows rather than a skills layer injected into an existing harness.Open source
OpenHands
You need a self-hosted autonomous agent platform with broader tool access beyond IDE-style coding.Open source
Cursor built-in agent mode
You want agent behavior without installing a third-party skills plugin, accepting that it lacks the enforced spec-first loop.Included with Cursor subscription
Claude Code native plugins
You prefer to compose your own lightweight prompt and skill set instead of adopting a full methodology.Free with Claude Code

What this trend reveals

Standardize AI agent behavior across a multi-harness team

Because Superpowers installs into ten harnesses with the same methodology, a team using Claude Code, Cursor, and Codex CLI can share one spec-first process instead of maintaining three prompt sets.

Confirm by installing on two harnesses and running the same feature spec through both; compare whether the planning and TDD steps match.

Reduce AI-generated rework by enforcing planning

The spec-chunk-then-plan-then-code loop targets the most expensive failure mode of AI coding: producing large volumes of code that miss the real requirement.

Measure rejected PRs or reverted commits before and after enabling Superpowers on a single project for two weeks.

Enterprise support path via PrimeRadiant

PrimeRadiant offers commercial support, additional tooling, and managed spending, which opens a procurement-friendly adoption route for organizations that cannot use unsupported open-source tooling alone.

Contact sales@primeradiant.com and confirm scope, SLA, and pricing before committing to a wider rollout.

Best next action

Install Superpowers on your primary harness and run one feature end-to-end

Pick the harness you use most — Claude Code via the official Anthropic marketplace is the fastest path — and take a small feature from spec to merged code using the default skills.

  1. In Claude Code, run `/plugin install superpowers@claude-plugins-official`.
  2. Restart the session and confirm the skills activate by starting a build-task conversation.
  3. Observe whether the agent asks clarifying spec questions before writing code.
  4. Approve the spec chunks, approve the implementation plan, and say 'go'.
  5. Verify the agent performs a red/green TDD cycle before delivering the implementation.

RepoDaily verdict

Superpowers is one of the few trending Shell projects that is not a dotfile or install script: it is a methodology runtime. Its value is the enforced spec → plan → TDD loop that runs across ten agent harnesses, and its credibility rests on MIT licensing, v6.1.0 maturity, and PrimeRadiant's commercial backing. Medium adoption risk comes from behavioral change and multi-harness maintenance, not licensing or security.

Sources