Superpower Agent
Adaptive PM -> Build -> QC -> Ops development system for Claude Code, OpenCode, Gemini CLI, Codex, Copilot, Cursor, Windsurf, and Antigravity.
superpower-agent gives teams a repeatable delivery operating system instead of ad-hoc prompting.
It is a custom FAD-branded system built by extracting and adapting the strongest ideas from:
This repo is not a fork. It repackages the useful patterns into one leaner install surface, one branded command namespace, and one stricter PM -> Build -> QC -> Ops pipeline.


Why This Exists
Most agent setups stop at prompt collections or copy entire upstream stacks into every project.
Superpower Agent takes a different approach:
- keep the PM -> Build -> QC -> Ops handoff explicit
- keep every major decision auditable
- keep brownfield risk and impact analysis mandatory
- keep Figma/Jira/Confluence/PR inputs first-class
- keep install size and context footprint under control
- keep one default surface:
/fad:*
Why Teams Use It
- Turn rough requirements into structured delivery artifacts.
- Enforce brownfield-safe coding decisions with risk/impact gates.
- Run one strict pipeline (
/fad:pipeline) with mandatory review and optimize phases.
- Keep one auditable flow from planning to deploy.
- Run consistent workflows across Claude Code, OpenCode, Gemini CLI, Codex, Copilot, Cursor, Windsurf, and Antigravity.
- Use branded command namespace:
/fad:*.
Install In 60 Seconds
npx superpower-agent init --dir /path/to/your-project
Bundle-aware install:
npx superpower-agent init --dir /path/to/your-project --bundle standard
npx superpower-agent init --dir /path/to/your-project --bundle core
npx superpower-agent init --dir /path/to/your-project --bundle full
During install, the CLI asks which runtime adapters to configure:
- Claude
- OpenCode
- Gemini CLI
- Codex
- Copilot
- Cursor
- Windsurf
- Antigravity
Non-interactive examples:
npx superpower-agent init --dir /path/to/your-project --all --no-prompt
npx superpower-agent init --dir /path/to/your-project --claude --opencode --gemini --codex --copilot --cursor --windsurf --antigravity --no-prompt
npx superpower-agent doctor --dir /path/to/your-project
npx superpower-agent estimate --bundle standard
npx superpower-agent inspect --dir /path/to/your-project
Command Namespace
- Primary namespace:
/fad:*
- Legacy
/gsd:* aliases remain available for migration only
full bundle restores the heavy legacy vendor tree from packaged archives on demand
Start here after install:
/fad:help
/fad:pipeline "<requirement or phase>"
What Gets Installed
CLAUDE.md
.claude/ workflows, scripts, rules, hooks, commands
.planning/ artifacts, install metadata, and audit scaffolding
.planning/setup/context-index.json for local context visibility
.claude-analysis/ kept as maintainer-only analysis context, not shipped in installed bundles
templates/vendor/*.tgz are used internally so npm publish stays lean while full still extracts legacy assets
- Runtime adapters:
.opencode/commands/fad-*.md (if OpenCode selected)
.gemini/commands/fad/*.toml (if Gemini CLI selected)
.codex/skills/fad-operator/SKILL.md (if Codex selected)
.github/copilot-instructions.md and .github/prompts/fad-*.prompt.md (if Copilot selected)
.cursor/rules/fad.mdc (if Cursor selected)
.windsurf/skills/fad-operator/SKILL.md and .windsurf/workflows/fad-*.md (if Windsurf selected)
.agent/skills/fad-operator/SKILL.md (if Antigravity selected)
AGENTS.md, GEMINI.md, CODEX.md bridge docs when relevant runtimes are selected
Optimization Highlights
- Bundle-aware install with
core, standard, and full
- Local context visibility with
superpower-agent estimate and superpower-agent inspect
- Project-local context index written to
.planning/setup/context-index.json
- Publish surface reduced from the raw vendor-tree approach to a lean npm artifact
- Runtime bridges are generated only for the runtimes you select at install time
full compatibility preserved through archive-backed extraction instead of shipping the entire legacy tree raw
Current packaging result after optimization:
- publish artifact:
178 files
- tarball size: about
3.7 MB
- unpacked package size: about
4.2 MB
full bundle still restores legacy assets when explicitly requested
Architecture Snapshot
Input (text / Jira / Confluence / Figma / PR)
-> /fad:pipeline
-> PM lane (intake, discovery, PRD, roadmap)
-> Build lane (plan to execution, fix loops)
-> Review lane (severity-first findings)
-> Optimize lane (maintainability/perf hardening)
-> Strict quality lane (lint/typecheck/test + security + risk)
-> Ops lane (health, deploy, incident, rollback)
\-> gates: risk + review + optimize + quality + security + health + design evidence
Runtime Support Matrix
| Claude | Native | .claude command contracts |
| OpenCode | Adapter | .opencode/commands/fad-*.md -> /fad-help, /fad-pipeline, ... |
| Gemini CLI | Adapter | .gemini/commands/fad/*.toml -> /fad:help, /fad:pipeline, ... |
| Codex | Adapter | .codex/skills/fad-operator/SKILL.md |
| Copilot | Adapter | .github/copilot-instructions.md + .github/prompts/fad-*.prompt.md |
| Cursor | Adapter | .cursor/rules/fad.mdc |
| Windsurf | Adapter | .windsurf/skills/fad-operator/SKILL.md + .windsurf/workflows/fad-*.md |
| Antigravity | Adapter | .agent/skills/fad-operator/SKILL.md + GEMINI.md bridge |
Professional Docs
References
Maintainers
npm run check
npm run vendor:refresh
npm run sync-template
npm run export-standalone -- /tmp/superpower-agent-repo
Publish
npm run check
npm publish --access public
For GitHub Actions publishing, set repository secret NPM_TOKEN or migrate the release workflow to npm trusted publishing.
License
MIT - see LICENSE.