skills for everythingusing-superpowers
Use when starting any conversation - establishes how to find and use skills, requiring Skill tool invocation before ANY response including clarifying questions
SKILL.md
| Name | using-superpowers |
| Description | Use when starting any conversation - establishes how to find and use skills, requiring Skill tool invocation before ANY response including clarifying questions |
Agentic-Skills
A multi-platform agentic ecosystem for AI-assisted development. It provides commands (or workflows), an orchestrator, specialist agents, and reusable skills so you can plan, implement, fix, test, deploy, and document projects in a consistent way across Cursor, Claude Code, and Google Anthropic Agent.
Quick Start
- Pick your platform — Cursor (
.cursor/commands/), Claude Code (.claude/commands/), or Google Anthropic Agent (.agent/workflows/). - Run a command — e.g.
/plan,/implement,/fix, or/docsfrom the command palette or chat. - Be specific — e.g.
/plan e-commerce site with cart and auth. The orchestrator routes to the right agents and skills.
What this repo is
- Entry points: 17 commands (Cursor/Claude) or workflows (Agent) that you invoke—e.g.
/plan,/implement,/fix,/docs. - Orchestrator: A single coordinator per platform that receives your request by mode, selects the right agents and skills, and runs the workflow. You don’t call agents or skills directly.
- Agents: Domain specialists (e.g. backend-specialist, frontend-specialist, documentation-writer, test-engineer) and process roles (e.g. project-planner, verifier). Each has its own skills and a domain color so the orchestrator (and logs) can see at a glance which area of responsibility is active.
- Skills: Reusable guidance and patterns (e.g. test-driven-development, writing-plans, frontend-design) used by the orchestrator and agents. Skills never call commands or the orchestrator.
Contract: You use only the entry points (commands or workflows). They route to the orchestrator; the orchestrator uses agents and skills. No cycles: skills don’t call commands or the orchestrator; agents don’t call commands.
Agent domains and colors
Agents are grouped by domain and share a basic color across all three platforms. This keeps roles non-overlapping and makes it easy for the orchestrator (and future colored logs) to see who owns what:
| Domain | Color | Cursor / Claude / Agent agents (examples) |
|---|---|---|
| Orchestration & planning | blue / violet | orchestrator, project-planner, product-manager |
| Discovery & legacy | cyan | explorer-agent, code-archaeologist |
| Verification & QA | amber | test-engineer, verifier (Cursor) |
| Security | red | security-auditor, penetration-tester |
| Backend & data | green | backend-specialist, database-architect |
| Frontend | indigo | frontend-specialist |
| DevOps & infra | slate | devops-engineer |
| Mobile | pink | mobile-developer |
| Documentation | zinc | documentation-writer |
| Performance | lime | performance-optimizer |
| SEO / marketing | yellow | seo-specialist |
| Games | emerald | game-developer |
On each platform, agents declare this in frontmatter (for example color: green on backend-specialist), and the orchestrator never asks an agent to work outside its domain.
Features
- Plan & create — Task breakdown, agent/skill assignments; new apps or modules from plan to implementation.
- Implement, fix & debug — Features and bugfixes with tests; systematic root-cause analysis.
- Test & review — Generate and run tests; multi-agent code review.
- Deploy & operate — Staging/production deployment; preview server and status.
- Document — Create or update docs in sync with the codebase.
- Explore & design — Brainstorm, enhance features, explain concepts, UI/UX design intelligence.
Platforms and entry points
| Platform | Entry path | How you start |
|---|---|---|
| Cursor | .cursor/commands/ | Run a command (e.g. /plan, /implement) from the Cursor command palette or chat. |
| Claude | .claude/commands/ | Run a command in Claude Code; it routes to the agent-orchestrator. |
| Agent (Google Anthropic) | .agent/workflows/ | Invoke a workflow; workflows are the entry point (no commands directory). |
Same 17 entry names on all platforms:
| Command / workflow | Typical use |
|---|---|
/plan | Create or refine a plan (task breakdown, agent/skill assignments). No code. |
/create | New app or major module; full lifecycle from plan to initial implementation. |
/implement | Implement features from an existing plan. |
/fix | Fix bugs and regressions with tests. |
/debug | Systematic debugging and root-cause analysis. |
/refactor | Refactor without changing behavior. |
/test | Generate and run tests, improve coverage. |
/review | Multi-agent code review. |
/docs | Create or update documentation (sync with code). |
/deploy | Coordinate deployment (staging/production). |
/status | Report project/agent/preview status (read-only). |
/preview | Manage preview server (start, stop, status, health). |
/brainstorm | Explore options before committing to implementation. |
/enhance | Add or update features in an existing app. |
/explain | Explain code or concepts (educator mode). |
/ui-ux-pro-max | Design intelligence (UI/UX, design system). |
/orchestrate | Multi-domain orchestration; orchestrator chooses agents/skills. |
You can chain modes (e.g. /brainstorm then /plan then /implement) for one task.
How to use the ecosystem
-
Start with a command (or workflow).
In Cursor or Claude, use the command (e.g./plan,/implement,/docs). On Agent, invoke the matching workflow. Do not call the orchestrator or agents directly. -
Be specific.
Example:/plan e-commerce site with cart and author/docs add README for the API module. The orchestrator will route to the right agents and skills. -
If your prompt has no command.
The workspace rule is: figure out which command(s) apply and mention them (e.g. “Use/planthen/implement”). Commands live in the platform’s commands (or workflows) directory. -
Plans.
Plans are written todocs/PLAN-{task-slug}.md. After/plan, use/createor/implementto execute; the orchestrator uses the plan to assign work. -
Documentation.
Use/docsfor anything doc-related; the orchestrator uses the documentation-writer agent and doc-related skills so docs stay aligned with the codebase.
Directory layout (per platform)
Layout is parallel across the three platforms; only the top-level folder and “commands vs workflows” differ.
| Directory | Purpose |
|---|---|
.cursor/ | Cursor IDE: commands, agents, rules, skills, scripts. |
.claude/ | Claude Code: commands, agents, skills, scripts. |
.agent/ | Google Anthropic Agent: workflows (no commands dir), agents, skills, scripts. |
docs/ | Project docs and plans (e.g. docs/PLAN-*.md). |
Under each platform:
commands/(orworkflows/for.agent/) — Entry points; each file has a Routing section to the orchestrator and a mode.agents/— Orchestrator plus specialist agents (e.g.orchestrator.md,documentation-writer.md,backend-specialist.md). Cursor also hasverifier.md.skills/— Reusable skills (e.g.documentation-templates,writing-plans,test-driven-development). Agents reference skills; skills do not call commands or the orchestrator.rules/(Cursor/Agent) — Workspace rules (e.g. entry-point, coding-style, git).scripts/— Helper scripts (e.g. preview, verification).
Validations
From the repo root, run ./run-validations.sh to execute link validation, dangling skills check, platform isolation, and docs secrets check. Reports are written to .reports/.
More detail
Orchestrator and modes: See the Command Modes (or Workflow Modes) table in the platform’s orchestrator:
.cursor/agents/orchestrator.md, .claude/agents/agent-orchestrator.md, .agent/agents/orchestrator.md.
- Skills: Browsable under
.cursor/skills/(and.claude/skills/,.agent/skills/). Each skill has aSKILL.mddescribing when and how it’s used.
Contributing
Open an issue or pull request on the repository.
License
See LICENSE.