agent-router

Files (4)

• SKILL.md
• 01-route-vague-idea.yaml · eval/cases/
• 02-route-bug.yaml · eval/cases/
• state.json · eval/state/

SKILL.md (first lines)

TOKEN CEILING: ~3K tokens. If skill exceeds, extract sections to references/.

# Agent Router — Bootstrap Skill

**This is the entry point for ALL work.** Load this skill first. It maps your request to the right agent + skill chain.

**User-invoked only** (`disable-model-invocation: true`) — zero context cost until you need it.

---

## Routing Table

| Your Request | Agent | Skills Loaded | Entry Point |
|--------------|-------|---------------|-------------|
| "I have an idea for..." / vague feature | `planner` → `product-thinking` → `planning-and-task-breakdown` → `dev-craft` | product-thinking, planning-and-task-breakdown, dev-craft | `skill("product-thinking")` |
| "Here's my requirements.xlsx" | `planner` → `project-discovery` → `planning-and-task-breakdown` → `dev-craft` | project-discovery, planning-and-task-breakdown, dev-craft | `skill("project-discovery")` |
| "Build the auth API per PLAN.md" | `implementer` → `dev-craft` | dev-craft, testing-strategies, code-review-and-quality | `skill("dev-craft")` |
| "Fix this failing test" | `debugger` → `prompt-optimizer` → `debugging-and-error-recovery` → `implementer` | prompt-optimizer, debugging-and-error-recovery, dev-craft, verification-before-completion | `skill("prompt-optimizer")` |
| "Review this PR" | `code-reviewer` → `prompt-optimizer` → `code-review-and-quality` → `verification-before-completion` | prompt-optimizer, code-review-and-quality, verification-before-completion | `skill("prompt-optimizer")` |
| "Security audit before launch" | `security-auditor` → `prompt-optimizer` → `bug-hunting` → `verification-before-completion` | prompt-optimizer, bug-hunting, code-review-and-quality, verification-before-completion | `skill("prompt-optimizer")` |
| "Audit agent architecture / diagnose wrapper layers" | `agent-architecture-audit` → `prompt-optimizer` → `verification-before-completion` | prompt-optimizer, verification-before-completion, debugging-and-error-recovery | `skill("prompt-optimizer")` |
| "Design the API for webhooks" | `api-designer` → `prompt-optimizer` → `api-design` | prompt-optimizer, api-design, dev-craft (CONTRACT phase) | `skill("prompt-optimizer")` |
| "Set up CI/CD for microservices" | `devops-engineer` → `prompt-optimizer` → `devops-automation` | prompt-optimizer, devops-automation, dev-craft | `skill("prompt-optimizer")` |
| "Weekly retrospective" | `retro-analyst` → `prompt-optimizer` → `retro` → `continuous-learning-v2` | prompt-optimizer, retro, continuous-learning-v2, context-engineering | `skill("prompt-optimizer")` |
| "Ship this release" | `shipper` → `prompt-optimizer` → `ship` → `verification-before-completion` | prompt-optimizer, ship, verification-before-completion | `skill("prompt-optimizer")` |
| "Optimize LLM costs" | `planner` → `cost-optimizer` | cost-optimizer | `skill("cost-optimizer")` |
| "My context is full / rotate session" | `context-guard` → `context-engineering` → `handoff` | context-engineering, handoff, continuous-learning-v2 | `skill("context-engineering")` |
| "What did we learn last sprint?" | `retro-analyst` → `prompt-optimizer` → `continuous-learning-v2` | prompt-optimizer, continuous-learning-v2, retro | `skill("prompt-optimizer")` |

---

## Comprehensive Routing Rules (from gstack)

### Product & Planning
- **New idea, brainstorming, "is this worth building", pitch a concept** → `/plan` (invokes `product-thinking` → `planning-and-task-breakdown`)
- **Spec something out, file an issue, write up a ticket, "turn this into a GitHub issue", "backlog item"** → `/spec` (invokes `project-discovery` → `product-thinking` → `planning-and-task-breakdown`)
- **Strategy, scope, ambition, "think bigger", "what should we build"** → `/plan` (invokes `grilling` → `architecture-decision-records`)
- **Architecture review, lock in the plan, "does this design make sense"** → `/review` (invokes `grilling` → `architecture-decision-records`)

### Development & Implementation
- **Build feature per PLAN.md** → `dev-craft` (or `ui-craft` for frontend)
- **Frontend/UI work, design system, component library** → `ui-craft`
- **API design, contract, OpenAPI, GraphQL schema** → `api-design` → `api-contract-designer`
- **Backend service, database, infrastructure** → `dev-craft` → `backend-patterns`, `database-migrations`
- **Mobile app, React Native, Expo** → `dev-craft` → `ui-craft` (mobile plugins)
- **What framework/library should I use?** → `tech-advisor` (research + recommend)
- **Is there a better alternative to [X]?** → `tech-advisor` (compare alternatives)
- **Should I upgrade [dependency]?** → `tech-advisor` (check versions + recommend)
- **What patterns does this codebase use?** → `ui-pattern-extractor` (extract existing patterns)
- **Build a component** → `ui-pattern-extractor` (first) → then `ui-component-builder`

### Debugging & Investigation
- **Bug, error, broken behavior, "why is this broken", "this doesn't work", "wtf", "something's wrong"** → `/investigate` (invokes `debugging-and-error-recovery` → `verification-before-completion`)
- **Failing test, build break, unexpected behavior** → `/investigate` (invokes `debugging-and-error-recovery`)
- **Performance issue, slow query, bottleneck** → `performance-profiler-and-tuner`

### Testing & Quality Assurance
- **Test the site, find bugs, QA, "does this work", "check the deploy"** → `/qa` (invokes `qa-and-edge-case-tester` → `testing-strategies` → `verification-before-completion`)
- **Just report bugs without fixing** → `qa-and-edge-case-tester` (analysis only)
- **Generate tests, edge cases, boundary testing** → `qa-and-edge-case-tester`
- **Visual regression, screenshot comparison** → `visual-regression`

### Code Review
- **Review code, check the diff, pre-landing review, "look at my changes"** → `/review` (invokes `code-review-and-quality` → `verification-before-completion`)
- **Security review, OWASP, vulnerabilities, "is this secure"** → `bug-hunting`

### Deployment & Shipping
- **Ship, deploy, push, create a PR, "let's land this", "send it"** → `/ship` (invokes `ship` → `verification-before-completion` → `verification-before-completion`)
- **Merge + deploy + verify as one flow** → `ship` (with full pipeline)
- **Configure deployment for the project** → `devops-automation`
- **Monitor prod after shipping, post-deploy checks** → `observability-engineering`
- **Update docs after shipping** → `documentation-engineering`

### Context & Session Management
- **Save progress, checkpoint, "save my work"** → `/context-save` (invokes `context-compressor-and-pruner` → `continuous-learning-v2` → `handoff`)
- **Resume, restore, "where was I", "continue work"** → `/context-restore` (invokes `context-engineering` → `handoff` → `continuous-learning-v2`)
- **Weekly retro, what did we ship, "how'd we do", "sprint retro"** → `/retro` (invokes `retro` → `continuous-learning-v2`)

### Documentation
- **Write docs from scratch, generate documentation, "document this feature/module"** → `documentation-engineering`
- **Update docs after shipping** → `documentation-engineering`

### Cost & Optimization
- **Optimize LLM costs, model routing, API budget, prompt caching** → `cost-optimizer`

---

## Quick Start

```
You: "Build a payment integration with Stripe"
Agent: loads agent-router → runs prompt-optimizer (pre-routing) → routes to planner
Planner: "I'll run product-thinking to refine, then planning-and-task-breakdown, then dev-craft"
```

```
You: "Review PR #247"
Agent: loads agent-router → runs prompt-optimizer (pre-routing) → routes to code-reviewer
Code Reviewer: "Running prompt-optimizer (per-agent) on the PR context, then code-review-and-quality with 8-axis review + pre-report gate"
```

```
You: "Debug the flaky login test"
Agent: loads agent-router → runs prompt-optimizer (pre-routing) → routes to debugger
Debugger: "Running prompt-optimizer (per-agent) on the bug report, then debugging-and-error-recovery 4-phase investigation"
```

---

## If Unsure: Triage First

If the request doesn't cleanly match a row above, the **Triage Agent** classifies it:

```
triage → classifies → routes to correct agent
```

Categories: Bug | Feature | Refactor | Security | Performance | Docs | Chore | Design | Test

---

From ad-free/agent-master-skills, MIT.

More from ad-free/agent-master-skills