app-platform-router（应用平台路由器）

App Platform Skills — Router

Overview

This skill routes DigitalOcean App Platform tasks to specialized sub-skills. Each sub-skill is optimized for a specific workflow, keeping context focused and responses precise.

Philosophy: These skills are opinionated playbooks, not documentation replicas. They make decisions for you (VPC by default, GitHub Actions for CI/CD, etc.) and only defer to docs for edge cases.

Key Principle: Skills contain only DO-specific knowledge the LLM doesn't have from training. Generic SQL, SDK patterns, and programming concepts are NOT duplicated here.

Trust the AI: The AI assistant should reason about the full workflow, chain skills together when needed, and adapt based on context (greenfield vs brownfield vs troubleshooting). Skills provide domain knowledge, not rigid scripts to follow blindly.

Available Skills

Development Phase

Architecture Phase

Operations Phase

Data Services

AI Services

Console SDK for AI Assistants

CRITICAL: AI assistants cannot use doctl apps console — it opens an interactive WebSocket session that requires human input. Use the do-app-sandbox Python package instead.

# Install
uv pip install do-app-sandbox  # or: pip install do-app-sandbox

Two Use Cases (Same Package, Different Skills)

# Debug existing app (troubleshooting skill)
app = Sandbox.get_from_id(app_id="your-app-id", component="web")

# Create new isolated sandbox (sandbox skill)
sandbox = Sandbox.create(image="python")

Full guide: See shared/console-sdk-patterns.md

Routing Decision Tree

┌─────────────────────────────────────────────────────────────────┐
│                        USER REQUEST                              │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
                ┌─────────────────────────┐
                │ Setting up development? │
                └─────────────────────────┘
                              │
                      "local" / "devcontainer"
                              ▼
                      ┌───────────┐
                      │devcontainers│
                      └───────────┘
                              │
                              ▼
                ┌─────────────────────────┐
                │ Designing or creating?  │
                └─────────────────────────┘
                      │           │           │
             "new app"│           │ "migrate" │ "plan" / "staged"
                      ▼           ▼           ▼
              ┌───────────┐  ┌───────────┐  ┌───────────┐
              │ designer  │  │ migration │  │  planner  │
              └───────────┘  └───────────┘  └───────────┘
                              │
                              ▼
                ┌─────────────────────────┐
                │    Shipping code?       │
                └─────────────────────────┘
                              │
                      "deploy" / "ship" / "release"
                              ▼
                      ┌───────────┐
                      │deployment │
                      └───────────┘
                              │
                              ▼
                ┌─────────────────────────┐
                │   Something broken?     │
                └─────────────────────────┘
                              │
                      "broken" / "failing" / "debug" / "logs"
                              ▼
                    ┌───────────────┐
                    │troubleshooting│
                    └───────────────┘
                              │
                              ▼
                ┌─────────────────────────┐
                │ Need isolated execution?│
                └─────────────────────────┘
                              │
                      "sandbox" / "isolated" / "execute code" / "code interpreter"
                              ▼
                      ┌───────────┐
                      │  sandbox  │
                      └───────────┘
                              │
                              ▼
                ┌─────────────────────────┐
                │ Configuring data?       │
                └─────────────────────────┘
                      │           │           │
           "postgres" │  "mysql"  │           │ "storage" / "S3" / "spaces"
           (complex)  │  "mongo"  │           │
                      │  "valkey" │           │
                      │  "kafka"  │           │
                      │  "opensearch"         │
                      ▼           ▼           ▼
              ┌───────────┐  ┌─────────────────────┐  ┌───────────┐
              │ postgres  │  │ managed-db-services │  │  spaces   │
              └───────────┘  └─────────────────────┘  └───────────┘
                              │
                              ▼
                ┌─────────────────────────┐
                │ AI inference needed?    │
                └─────────────────────────┘
                              │
                      "gradient" / "LLM" / "inference"
                              ▼
                      ┌─────────────┐
                      │ ai-services │
                      └─────────────┘

Workflow Chaining

The AI assistant should reason about the full workflow, not just route to a single skill.

Workflow Types

Chaining Logic

┌─────────────────────────────────────────────────────────────────┐
│                    UNDERSTAND THE GOAL                           │
│  Is this a single task or a multi-phase workflow?                │
└─────────────────────────────────────────────────────────────────┘
                              │
              ┌───────────────┴───────────────┐
              │                               │
              ▼                               ▼
       ┌─────────────┐                 ┌─────────────┐
       │ Single Task │                 │  Workflow   │
       │             │                 │             │
       │ Route to    │                 │ Chain       │
       │ one skill   │                 │ multiple    │
       └─────────────┘                 └─────────────┘
              │                               │
              ▼                               ▼
       troubleshooting              migration → designer →
       postgres                     planner → deployment
       spaces
       etc.

Examples

"I want to build a new event-driven app with Kafka"

1. Start with designer → create .do/app.yaml with Kafka
2. Use managed-db-services for Kafka-specific bindable variables
3. Use planner → generate staged Plan/ files
4. Execute with deployment skill

"Migrate my Heroku app to DigitalOcean"

1. Start with migration → Heroku deep chapter auto-routes based on intent (Q&A, Guided, Auto-Migrate)
2. Parses Procfile, app.json, heroku.yml, maps add-ons, generates .do/app.yaml
3. Use planner → plan the staged deployment
4. Execute with deployment skill

"My deployed app keeps returning 502"

1. Route directly to troubleshooting (single skill, no chaining)
2. Use debug container, analyze logs, fix issue

"Add PostgreSQL to my existing app"

1. Use postgres for schema/user setup
2. Update .do/app.yaml with database binding
3. Redeploy with deployment skill

Key Principle

The AI assistant should:

1. Understand the overall goal (not just the immediate request)
2. Identify which skills are needed and in what order
3. Chain skills together, passing artifacts between them
4. Adapt based on what already exists (greenfield vs brownfield)

Trust the AI to reason. These skills provide domain knowledge, not rigid scripts.

Trigger Phrases Reference

Third-Party Integrations (No Skill Needed)

For integrations not covered by dedicated skills (DataDog, Sentry, New Relic, Stripe, etc.):

1. Get credentials from the vendor
2. Add to GitHub Secrets (Repo → Settings → Secrets → Actions)
3. Reference in app spec: envs: - key: DATADOG_API_KEY scope: RUN_TIME value: ${DATADOG_API_KEY} # From GitHub Secrets
4. Follow vendor documentation for SDK/agent setup

→ The agent's training covers vendor-specific patterns — no skill required.

Credential Handling Philosophy

CRITICAL: These skills are designed to keep credentials out of agent hands.

Priority Order (Most to Least Preferred)

1. GITHUB SECRETS (RECOMMENDED DEFAULT)
   ├── Agent never sees credentials
   ├── User manually adds: Repo → Settings → Secrets → Actions
   ├── Workflow references: ${{ secrets.DATABASE_URL }}
   └── Agent generates workflow, user handles secrets

2. APP PLATFORM BINDABLE VARIABLES
   ├── For DO Managed Databases (Postgres, MySQL, MongoDB, etc.)
   ├── Credentials injected via ${db.DATABASE_URL}
   ├── Agent configures app spec, never sees credentials
   └── Best for: Apps using DO managed services

3. LOCAL .ENV + EPHEMERAL APP SPEC
   ├── User maintains .env file locally
   ├── Agent creates temp app spec with placeholders
   └── Substitutes values → Deploys → Deletes temp file

4. EXTERNAL SERVICES
   ├── Same patterns as #1 or #3
   └── User responsible for credential management

Artifact Contracts

All skills produce and consume artifacts with consistent naming:

Handoff Protocol

When a skill completes, it should:

1. State the artifact(s) produced: "Created .do/app.yaml with 3-component architecture"
2. Indicate the file path: Relative to project root
3. Suggest next skill if applicable: "To deploy this, use the deployment skill"

Plugin/Tool Requirements

Sub-Skill Locations

app-platform-skills/
├── SKILL.md                              # This file (router)
├── skills/
│   ├── devcontainers/SKILL.md
│   ├── designer/SKILL.md
│   ├── migration/SKILL.md
│   ├── deployment/SKILL.md
│   ├── planner/SKILL.md                  # Staged project plans (design to deployment)
│   │   └── templates/                    # Local + Tier 1, 2, 3 stage templates
│   ├── troubleshooting/SKILL.md
│   ├── sandbox/SKILL.md                  # Isolated containers for AI agent execution
│   ├── postgres/SKILL.md
│   ├── managed-db-services/SKILL.md      # MySQL, MongoDB, Valkey, Kafka, OpenSearch
│   ├── spaces/SKILL.md
│   └── ai-services/SKILL.md              # Gradient AI
└── shared/
    ├── app-spec-schema.yaml
    ├── artifact-contracts.md
    └── credential-patterns.md

Opinionated Defaults Summary

Escalation

If a task doesn't fit any skill or spans multiple skills ambiguously:

1. Ask clarifying question to determine primary intent
2. Suggest skill sequence if task is multi-phase
3. For truly novel tasks, fall back to App Platform documentation: https://docs.digitalocean.com/products/app-platform/