skillsSource-backedReview first Safety ✓ Privacy ✓
Superpowers Skills
MIT-licensed Superpowers skill and plugin framework by Jesse Vincent for Claude Code, Codex App, Codex CLI, Cursor, Gemini CLI, Antigravity, Kimi Code, OpenCode, Pi, GitHub Copilot CLI, and other coding agents, covering brainstorming, planning, TDD, systematic debugging, subagent-driven development, code review, git worktrees, and finish-the-branch workflows.
HarnessClaude CodeCodexWindsurfGeminiCursorCLI
Level:expertType:capability-packVerified:validated
Review first — review before installing
Open the source and read safety notes before installing.
Safety notes
- Superpowers installs skills plus harness-specific bootstrap or hook behavior that can affect how an agent responds from session start. Review installed hooks and plugin metadata for the target harness.
- The `using-superpowers` skill strongly requires skill checks before agent responses, while also stating that explicit user and project instructions take precedence. Keep that priority order intact.
- The workflow skills can direct agents to create specs, plans, branches, worktrees, tests, commits, subagent tasks, review packages, and long-running implementation loops.
- Subagent-driven development can run for extended periods and dispatch multiple agents. Use clear budgets, model selection rules, task boundaries, and stop conditions.
- The TDD skill intentionally requires failing tests before production code. Confirm that this discipline fits the project before enabling it as a default workflow.
- The optional visual companion uses a browser/server flow during brainstorming. Review local server behavior, ports, and auth before using it with private project context.
Privacy notes
- Superpowers workflows can expose product ideas, specs, design docs, implementation plans, source code, tests, diffs, review findings, git history, branch names, tool outputs, and agent handoff prompts.
- The README states that the optional visual companion may load the Prime Radiant logo from the creator's website with the Superpowers version, and can be disabled with `SUPERPOWERS_DISABLE_TELEMETRY` or compatible Claude telemetry opt-outs.
- Do not include secrets, customer data, unpublished product strategy, private incidents, or proprietary code in public examples, review packages, support issues, or visual companion artifacts.
- Subagent prompts and review packages should be treated as private development artifacts because they may include source snippets, diffs, file paths, test output, and architecture decisions.
Prerequisites
- A supported coding-agent harness and its plugin or extension install path.
- A repository where Superpowers can add skills, startup hooks, and workflow instructions for the selected agent.
- A willingness to follow structured workflows such as brainstorming, planning, TDD, subagent implementation, code review, and branch finishing.
- Project-specific instructions that clearly state where Superpowers workflows should be adapted or overridden.
- Telemetry and network policy review before using the optional brainstorming visual companion.
Schema details
- Install type
- package
- Reading time
- 7 min
- Difficulty score
- 87
- Troubleshooting
- Yes
- Breaking changes
- No
Source repository stats
- Scope
- Source repo
Package metadata
Skill and platform metadata
- Skill type
- capability-pack
- Skill level
- expert
- Verification
- validated
- Verified at
- 2026-06-18
Retrieval sources
https://github.com/obra/superpowers/blob/main/README.mdhttps://github.com/obra/superpowers/blob/main/.claude-plugin/plugin.jsonhttps://github.com/obra/superpowers/blob/main/.codex-plugin/plugin.jsonhttps://github.com/obra/superpowers/blob/main/hooks/hooks-codex.jsonhttps://github.com/obra/superpowers/blob/main/skills/using-superpowers/SKILL.mdhttps://github.com/obra/superpowers/blob/main/skills/subagent-driven-development/SKILL.mdhttps://github.com/obra/superpowers/blob/main/skills/brainstorming/SKILL.md
Tested platforms
Claude CodeCodex AppCodex CLICursorGemini CLIAntigravityKimi CodeOpenCodePiGitHub Copilot CLI
| Platform | Support | Install path |
|---|---|---|
| claude-code | Native | .claude/skills/<skill-name>/SKILL.md |
| codex | Native | .agents/skills/<skill-name>/SKILL.md |
| windsurf | Native | .windsurf/skills/<skill-name>/SKILL.md |
| gemini | Native | .gemini/skills/<skill-name>/SKILL.md or .agents/skills/<skill-name>/SKILL.md |
| cursor | Adapter | .cursor/rules/<skill-name>.mdc |
| cli | Manual | AGENTS.md or tool-specific context file |
Full copyable content
/plugin install superpowers@claude-plugins-official
# Codex users can install Superpowers from the official Codex plugin
# marketplace through the app or `/plugins` flow.About this resource
Superpowers Skills
Superpowers is a coding-agent skills framework and software development methodology. It packages skills, plugin manifests, startup hooks, and cross-harness instructions so agents follow structured workflows for design, planning, TDD, debugging, subagent execution, code review, worktrees, and branch completion.
Use this listing for the Superpowers skill/plugin pack. Use separate entries when evaluating an individual agent harness, a code-review tool, or a specific subagent runtime.
Knowledge Freshness
Verified on 2026-06-18, obra/superpowers reported version 6.0.2 in
its Claude and Codex plugin metadata, had a latest GitHub release v6.0.2
published on 2026-06-17, and showed active repository updates on 2026-06-18.
The repository supports multiple agent harnesses, and each harness has its own install path. Re-check the README and plugin metadata before installing, because marketplace behavior, hooks, and update paths can differ by host.
Retrieval Sources
This listing is grounded in:
- The upstream README and current GitHub repository metadata.
- Claude Code and Codex plugin manifests.
- The Codex hook configuration.
- Representative
using-superpowers,subagent-driven-development, andbrainstormingskill files. - The latest GitHub release metadata and MIT license metadata.
Core Workflow
Claude Code users can install from the official Claude plugin marketplace:
/plugin install superpowers@claude-plugins-official
Codex App and Codex CLI users install from the official Codex plugin
marketplace through the app or /plugins flow. Other supported hosts use
their own install commands, including Cursor, Gemini CLI, Antigravity,
OpenCode, Kimi Code, Pi, Factory Droid, and GitHub Copilot CLI.
Once installed, Superpowers uses startup/bootstrap behavior so the agent sees the skill library and checks for relevant skills before coding-agent work.
Capability Scope
| Area | Coverage |
|---|---|
| Design and planning | Brainstorming, spec writing, implementation planning, and plan execution workflows |
| Implementation discipline | Test-driven development, verification before completion, and systematic debugging |
| Subagents | Subagent-driven development, parallel-agent dispatch, task review, review packages, and model selection guidance |
| Code review | Requesting and receiving code review, severity-based review findings, and finish-the-branch decisions |
| Git workflows | Git worktrees, branch isolation, task commits, final review, merge/PR/keep/discard options |
| Skill authoring | Writing and testing new skills, including examples and skill quality guidance |
| Cross-harness support | Claude Code, Codex, Cursor, Gemini CLI, Antigravity, Kimi, OpenCode, Pi, Copilot CLI, and related plugin metadata |
Use Cases
- Add a structured software-development workflow to an AI coding agent.
- Make an agent slow down for brainstorming and design before implementation.
- Enforce red-green-refactor TDD on feature and bug-fix work.
- Run implementation plans through isolated subagents and task-scoped review.
- Improve debugging by requiring root-cause tracing and verification before declaring completion.
- Coordinate branch finishing, final review, and PR or merge decisions.
Production Rules
- Confirm the installed host and marketplace path before relying on a Superpowers workflow.
- Keep explicit user and project instructions higher priority than Superpowers defaults.
- Review session-start hooks and bootstrap files before enabling the plugin in a private or regulated codebase.
- Do not let subagent loops run without budgets, task boundaries, model selection rules, and clear stop conditions.
- Treat generated specs, plans, review packages, and subagent prompts as private development artifacts.
- Disable optional telemetry or visual companion network access when project policy requires local-only behavior.
Source Review
Verified on 2026-06-18:
- GitHub metadata reported
obra/superpowersas an MIT-licensed repository with more than 200,000 stars, default branchmain, latest releasev6.0.2, and active updates on 2026-06-18. - The README described Superpowers as a software development methodology for coding agents with supported install paths for Claude Code, Antigravity, Codex App, Codex CLI, Cursor, Factory Droid, Gemini CLI, GitHub Copilot CLI, Kimi Code, OpenCode, and Pi.
.claude-plugin/plugin.jsonand.codex-plugin/plugin.jsonboth declared version6.0.2, MIT licensing, and skills/workflow coverage for planning, TDD, debugging, collaboration, code review, and agent workflows.hooks/hooks-codex.jsondeclared a CodexSessionStarthook that runs the Superpowers session-start bootstrap for startup, resume, and clear events.skills/using-superpowers/SKILL.mddescribed skill invocation rules, platform adaptation, user-instruction precedence, and skill-loading mechanisms across Claude Code, Codex, Copilot CLI, Gemini CLI, and other environments.skills/subagent-driven-development/SKILL.mddescribed fresh implementer subagents, per-task review, final branch review, model selection, and stop conditions for plan execution.skills/brainstorming/SKILL.mddescribed the design-before-implementation workflow, approval gates, spec writing, review, and optional visual companion behavior.
Source citations
Add this badge to your README
Show that Superpowers Skills is listed on HeyClaude. Paste this Markdown into your README — it renders the badge and links back to this page.
[](https://heyclau.de/entry/skills/superpowers-skills)How it compares
Superpowers Skills side by side with 3 alternatives on trust, install, platform support, and disclosed safety notes — all from reviewed registry metadata.
| Field | MIT-licensed Superpowers skill and plugin framework by Jesse Vincent for Claude Code, Codex App, Codex CLI, Cursor, Gemini CLI, Antigravity, Kimi Code, OpenCode, Pi, GitHub Copilot CLI, and other coding agents, covering brainstorming, planning, TDD, systematic debugging, subagent-driven development, code review, git worktrees, and finish-the-branch workflows. Open dossier | Addy Osmani's production-grade Agent Skills pack for AI coding agents, with lifecycle slash commands, engineering workflow skills, review personas, quality gates, and cross-agent setup guidance for Claude Code, Cursor, Gemini CLI, Antigravity CLI, OpenCode, GitHub Copilot, and other agents. Open dossier | MIT-licensed BrowserAct Agent Skill pack for installing and operating the `browser-act` browser automation CLI from Claude Code, Codex, OpenClaw, Cursor, OpenCode, Windsurf, Gemini CLI, and other skills-compatible agents. Open dossier | Official Hugging Face Agent Skills collection for Claude Code, Codex, Cursor, Gemini CLI, and other skills-compatible agents, covering Hub CLI workflows, datasets, model search, Spaces, Gradio, fine-tuning, evaluations, local models, papers, Trackio, ZeroGPU, transformers.js, TRL, and the Hugging Face MCP server. Open dossier |
|---|---|---|---|---|
| Trust | ||||
| Install risk | Review first | Review first | Review first | Review first |
| Notes | Safety ✓ Privacy ✓ | Safety ✓ Privacy ✓ | Safety ✓ Privacy ✓ | Safety ✓ Privacy ✓ |
| Brand | ||||
| Category | skills | skills | skills | skills |
| Source | source-backed | source-backed | source-backed | source-backed |
| Author | Jesse Vincent | Addy Osmani | BrowserAct | Hugging Face |
| Added | 2026-06-18 | 2026-06-18 | 2026-06-18 | 2026-06-18 |
| Platforms | Claude CodeCodexWindsurfGeminiCursorCLI | Claude CodeCodexWindsurfGeminiCursorCLI | Claude CodeCodexWindsurfGeminiCursorCLIVS Code | Claude CodeCodexWindsurfGeminiCursorCLI |
| Source repo | — | — | — | — |
| Safety notes | ✓Superpowers installs skills plus harness-specific bootstrap or hook behavior that can affect how an agent responds from session start. Review installed hooks and plugin metadata for the target harness. The `using-superpowers` skill strongly requires skill checks before agent responses, while also stating that explicit user and project instructions take precedence. Keep that priority order intact. The workflow skills can direct agents to create specs, plans, branches, worktrees, tests, commits, subagent tasks, review packages, and long-running implementation loops. Subagent-driven development can run for extended periods and dispatch multiple agents. Use clear budgets, model selection rules, task boundaries, and stop conditions. The TDD skill intentionally requires failing tests before production code. Confirm that this discipline fits the project before enabling it as a default workflow. The optional visual companion uses a browser/server flow during brainstorming. Review local server behavior, ports, and auth before using it with private project context. | ✓The slash commands are designed to guide real coding, testing, reviewing, committing, and shipping work; keep edits, commits, pushes, CI changes, and deploys behind the host's normal approval controls. `/build auto` is explicitly intended to generate a plan and implement multiple tasks in one approved pass. Use it on bounded specs, review the generated plan first, and stop on test failures or risky changes. The skills encode durable engineering workflows, not guaranteed-current framework APIs. Follow the source-driven-development guidance and verify current documentation before applying generated code. Security, CI/CD, observability, migration, and launch skills can touch production-sensitive systems. Require dry-run plans, rollback notes, and environment scoping before approving operational commands. Review personas and quality gates are useful second opinions, but they do not replace maintainer review, domain-specific tests, threat modeling, or release sign-off. | ✓BrowserAct can open pages, click, type, upload files, inspect state, capture screenshots, read page text, handle dialogs, export cookies, capture network requests, and operate logged-in browser sessions. Use BrowserAct only on sites, accounts, and data sources where the user has authorization. Do not use it to evade access controls, violate site terms, scrape disallowed data, or bypass rate limits. The entry skill declares confirmation gates for browser creation, deletion, local Chrome profile import, proxy/security changes, logins, form submissions, file uploads, and other sensitive operations; preserve those gates in agent workflows. `solve-captcha` may send the challenge image to BrowserAct's verification-assistance service according to the skill metadata; do not use it with sensitive or unauthorized pages. `remote-assist` can generate a live handoff URL for a human to take over. Treat that URL as access to the active browser session. Skill Forge can generate reusable automation skills from explored sites. Review generated scripts, selectors, network assumptions, output schemas, and site authorization before reusing them at scale. | ✓Hugging Face Skills can guide agents through Hub reads and writes, dataset uploads, model publishing, Space creation, training jobs, evaluation runs, repo settings, discussions, pull requests, secrets, variables, webhooks, and endpoint operations. Keep destructive or billable operations behind explicit approval: repo deletion, file deletion, private-to-public changes, endpoint deployment, hardware upgrades, Spaces volume changes, webhook creation, and cloud Job submission. Prefer read-only model, dataset, paper, and Space discovery before allowing write actions. Use dry-run modes when available for uploads, syncs, cache cleanup, dataset extraction, and infrastructure changes. The Hugging Face MCP server can search Hub assets, fetch docs, invoke MCP-enabled Gradio Spaces, and run compute jobs. Treat Space invocations and returned content as untrusted third-party tool output. Training and fine-tuning skills can consume paid GPU time and write models to the Hub. Validate datasets, model licenses, output visibility, timeout settings, and token scope before starting jobs. Do not publish generated model cards, datasets, papers, traces, or Spaces until licenses, attribution, evaluation claims, safety notes, and privacy constraints have been reviewed. |
| Privacy notes | ✓Superpowers workflows can expose product ideas, specs, design docs, implementation plans, source code, tests, diffs, review findings, git history, branch names, tool outputs, and agent handoff prompts. The README states that the optional visual companion may load the Prime Radiant logo from the creator's website with the Superpowers version, and can be disabled with `SUPERPOWERS_DISABLE_TELEMETRY` or compatible Claude telemetry opt-outs. Do not include secrets, customer data, unpublished product strategy, private incidents, or proprietary code in public examples, review packages, support issues, or visual companion artifacts. Subagent prompts and review packages should be treated as private development artifacts because they may include source snippets, diffs, file paths, test output, and architecture decisions. | ✓Using the pack with an AI agent can expose repository code, product requirements, architecture notes, tests, CI logs, deployment settings, incidents, security findings, and launch plans to the configured model provider. Do not paste secrets, customer data, private incident records, production credentials, unpublished roadmap details, or proprietary compliance material into public prompts, issues, screenshots, or PR bodies. Agent personas and review workflows may ask for browser traces, performance data, logs, build output, dependency lists, and environment details; redact tokens and private URLs before sharing artifacts. | ✓BrowserAct workflows can expose page content, screenshots, URLs, credentials typed into forms, cookies, browser profiles, uploaded files, downloaded files, network requests, HAR data, session names, browser descriptions, and logs. The BrowserAct skill metadata states that cookies, login sessions, page content, credentials, and browser profile data stay local, except the CAPTCHA challenge image when `solve-captcha` is invoked. Chrome-direct and profile import workflows can connect agents to existing local browser state. Treat those modes as account access, not a blank test browser. Log reports, feedback, Discord support, generated Skill Forge packages, and shared screenshots can leak private browsing or account context if submitted without review. Managed proxy, stealth browser, and API-key features create additional BrowserAct service dependencies beyond local CLI execution. | ✓Hub workflows can expose `HF_TOKEN`, private model or dataset names, training data, evaluation prompts, model outputs, papers, local file paths, logs, traces, secrets, Space variables, endpoint configuration, and organization membership. Agent trace upload workflows should default to private dataset repos because traces may include prompts, source code, tool output, file paths, credentials, screenshots, personal data, or customer context. Dataset Viewer, MCP, Jobs, Spaces, Inference Endpoints, Gradio apps, and third-party model repositories may receive user queries, files, prompts, examples, and generated outputs. Use least-privilege tokens, avoid passing tokens directly in command arguments when environment variables are supported, and redact logs before sharing PRs, issues, screenshots, or support requests. Check model, dataset, and Space licenses before using downloaded assets for training, redistribution, commercial work, or public demos. |
| Prerequisites |
|
|
|
|
| Install | | | | |
| Config | — | — | — | — |
| Citations | ||||
| Claim | Unclaimed | Unclaimed | Unclaimed | Unclaimed |
Related resources
Alternatives
Featured in
Signals
Loading live community signals…
More like this, weekly
A short, calm digest of reviewed Claude resources. Unsubscribe any time.