mtsbatalha/skill-router

Skill Router

When to Use

Use this skill when:

• The user says "I don't know where to start" or "which skill should I use"
• The user has a vague goal without a clear method
• The user asks "what should I use for..." or "I'm not sure how to approach this"
• The user is new to the skill library and needs guidance

Goal

Help users who are unsure of what they want to do or which skill to use. Interview them with a short structured conversation, then recommend the most relevant skill(s) from the installed library — with a clear explanation of why each skill fits and exactly how to invoke it.

---

Instructions

Step 1 — Acknowledge and open the interview

Respond warmly and tell the user you'll ask a few quick questions to find the right skill for them. Do NOT suggest any skills yet.

Example opener:

"No problem — let me ask you a few quick questions so I can point you to exactly the right skill."

---

Step 2 — Ask the Funnel Questions (one at a time, in order)

Ask only what you need. If an earlier answer makes a later question irrelevant, skip it.

Q1 — What is the broad area of the task? Present these as numbered options:

1. Building / coding something (app, feature, component, script)
2. Fixing or debugging something that's broken
3. Security, pentesting, or vulnerability assessment
4. AI agents, LLMs, or automation pipelines
5. Marketing, SEO, content, or growth
6. DevOps, infrastructure, deployment, or git
7. Design, UI/UX, or creative output
8. Planning, strategy, or documentation
9. Something else (ask them to describe it)

Q2 — How specific is the task?

1. I have a clear spec / I know exactly what I want built
2. I have a rough idea but need help shaping it
3. I'm totally starting from scratch with no clear direction

Q3 — What tech stack or domain is involved? (only ask if relevant) Examples: React / Next.js, Node.js, Python, AWS, Stripe, AI/LLM, no-code, etc. If they say "not sure" or "any", that's fine — move on.

Q4 — Do you want to work autonomously (agent does everything) or collaboratively (you stay in the loop)?

1. Fully autonomous — just go
2. Collaborative — I want to review/approve steps
3. Not sure yet

---

Step 3 — Recommend skills

Based on their answers, recommend 1 primary skill and up to 2 secondary skills. Structure your recommendation exactly like this:

✅ Primary Skill: @skill-name Why: [1–2 sentences explaining why this is the best fit for what they described] Invoke it like this:

@skill-name [paste their goal here]

🔁 Also consider:

• @skill-name-2 — [one sentence on when to layer this in]
• @skill-name-3 — [one sentence on when to layer this in]

---

Step 4 — Offer a ready-made prompt

After the recommendation, offer to write them a ready-to-use prompt they can paste directly into Antigravity. Ask:

"Want me to write the full prompt for you so you can just paste it in?"

If yes, compose a complete, specific prompt using the recommended skill and everything you learned about their goal from the interview.

---

Example Interaction

User: @skill-router I want to build something but I'm not sure where to start

Agent: No problem — let me ask a few quick questions.
       What is the broad area?
       (1) Building  (2) Debugging  (3) Security  (4) AI/LLM
       (5) Marketing  (6) DevOps  (7) Design  (8) Planning

User: 1 - building something

Agent: How specific is the task?
       (1) Clear spec  (2) Rough idea  (3) Starting from scratch

User: 2 - rough idea

Agent: What tech stack? (React, Node, Python, etc. — or "not sure")

User: Next.js

Agent: ✅ Primary Skill: @brainstorming
       Why: You have a rough idea that needs shaping before building.
       Brainstorming asks structured questions and produces a clear spec.

       @brainstorming help me design a [your app idea] using Next.js

       🔁 Also consider:
       - @plan-writing — once brainstorming produces a spec, break it into tasks
       - @senior-fullstack — when you are ready to start building

       Want me to write the full prompt for you?

---

Skill Routing Reference

Building a full product or app from scratch

• Primary: @app-builder
• If they want to plan first: @brainstorming → @plan-writing → @app-builder
• If they want it fully autonomous: @loki-mode

Building a specific frontend feature / UI

• Primary: @senior-fullstack or @frontend-design
• Stack-specific: @react-patterns, @nextjs-best-practices, @tailwind-patterns
• If they want a full design system: @ui-ux-pro-max + @core-components

Building a backend API or service

• Primary: @backend-dev-guidelines
• Stack-specific: @nodejs-best-practices, @python-patterns, @nestjs-expert
• API design: @api-patterns
• Database: @database-design + @prisma-expert

Debugging something broken

• Primary: @systematic-debugging
• If tests are failing: @test-fixing
• If it's a code quality issue: @clean-code

Writing tests / TDD

• Primary: @tdd
• For Playwright/browser tests: @playwright-skill
• For Jest patterns: @testing-patterns

Integrating a third-party service

• Payments: @stripe-integration
• Auth: @clerk-auth or @nextjs-supabase-auth
• Database: @neon-postgres or @firebase
• Messaging: @twilio-communications
• Bots: @slack-bot-builder, @discord-bot-architect, @telegram-bot-builder
• File storage: @file-uploads
• Analytics: @analytics-tracking

AI / LLM / agents

• Architecture: @ai-agents-architect
• RAG pipelines: @rag-engineer
• Prompts: @prompt-engineer
• Multi-agent: @langgraph or @crewai
• Observability: @langfuse
• Voice: @voice-agents

Security / pentesting

• Start here: @ethical-hacking-methodology + @pentest-checklist
• Web app testing: @burp-suite-testing, @sql-injection-testing, @xss-html-injection
• Network/infra: @aws-penetration-testing, @linux-privilege-escalation
• Reference: @top-web-vulnerabilities

DevOps / infrastructure / deployment

• Docker: @docker-expert
• Cloud: @aws-serverless, @gcp-cloud-run, @vercel-deployment
• Git workflow: @git-pushing, @using-git-worktrees, @github-workflow-automation
• Scripting: @linux-shell-scripting

Marketing / growth / SEO

• Copy: @copywriting
• Landing pages: @page-cro
• SEO: @seo-fundamentals + @seo-audit
• Email: @email-sequence
• Ads: @paid-ads
• Launch: @launch-strategy

Planning / architecture / strategy

• Quick plan: @concise-planning
• Full plan: @plan-writing → @executing-plans
• Architecture: @software-architecture or @senior-architect
• Product strategy: @product-manager-toolkit

Creative / design / visuals

• UI: @frontend-design
• Data viz: @claude-d3js-skill
• Generative art: @algorithmic-art
• Presentations: @pptx-official

Fully autonomous / parallel execution

• Full startup mode: @loki-mode
• Independent parallel tasks: @dispatching-parallel-agents
• Plan then execute: @subagent-driven-development

Document creation

• Word doc: @docx-official
• PDF: @pdf-official
• Spreadsheet: @xlsx-official
• Presentation: @pptx-official

---

Constraints

• Never recommend more than 1 primary skill and 2 secondary skills at a time.
• Always include the exact @invoke syntax so users can copy-paste it.
• If the user's goal spans multiple categories, pick the most upstream skill (e.g. @brainstorming before @senior-fullstack).
• Do not overwhelm the user with the full skill list. Recommend only what is relevant to their specific answers.
• If the user is totally lost, default to @brainstorming for open-ended goals, or @app-builder for anything involving building something.
• After recommending, always offer to write a ready-made prompt for them.

---

Limitations

• Only recommends skills from the installed library. If a skill is not installed, the recommendation may not work.
• Routing is based on natural language matching. Highly ambiguous goals may require follow-up clarification.
• Does not execute the recommended skill — it only recommends it. The user must invoke the skill themselves.
• The routing reference covers the most common skills but does not include every skill in the library.