Skill

Claude Code Autodeploy Skill

Purpose

This skill teaches Claude Code how to safely deploy applications to Coolify or Dokploy by:

1. Inspecting a local project to understand its structure
2. Inferring deployment requirements from project files
3. Generating a deploy.intent.yaml manifest
4. Selecting the appropriate deployment platform
5. Executing MCP tool calls in safe order
6. Verifying deployment success and reporting results

Supported Platforms

Both platforms support:

• Git-based source deployments (nixpacks, dockerfile)
• Prebuilt Docker images
• Docker Compose stacks
• Custom domains with SSL
• Environment variables and secrets
• Health checks

How Claude Should Consume This Skill

When Claude Receives a Deployment Request

1. Acknowledge the request and ask clarifying questions if needed
2. Inspect the local project (never deploy blind)
3. Analyze project files to determine requirements
4. Generate a deploy.intent.yaml manifest
5. Select platform based on decision criteria
6. Plan the deployment steps
7. Execute MCP calls in correct order
8. Verify deployment health
9. Report results to user

Skill Invocation

This skill is automatically available to Claude Code. No explicit invocation needed.

Claude should follow the instructions in this skill whenever the user asks to:

• "deploy this app"
• "set up CI/CD"
• "push to production"
• or uses any deployment-related commands

What This Skill Expects

From the User

From the Project

Claude will inspect these files:

What This Skill Produces

Primary Output: deploy.intent.yaml

A normalized deployment manifest that describes:

• Project metadata (name, version, owner)
• Repository source (provider, repo, branch)
• Runtime (language, version, package manager)
• Build strategy (nixpacks, dockerfile, prebuilt)
• Runtime config (port, health check, resources)
• Domains (FQDN, HTTPS settings)
• Environment variables
• Secrets placeholders
• Deployment targets (coolify/dokploy)

Secondary Outputs

1. MCP Tool Calls: Sequenced API calls to the chosen platform
2. Deployment Status: Success/failure with logs
3. Verification Results: Health check outcomes

Core Principles

Never Deploy Blindly

Claude MUST:

• ✅ Inspect the project before any deployment action
• ✅ Read package.json, Dockerfile, compose files
• ✅ Identify exposed ports from code and config
• ✅ Check for required env vars in documentation
• ✅ Warn about missing secrets before deploying

Claude MUST NOT:

• ❌ Trigger deployment without inspecting project
• ❌ Assume default ports (e.g., always 3000)
• ❌ Skip health check verification
• ❌ Deploy with known missing secrets

Prefer Explicit Configuration

Claude SHOULD:

• ✅ Use explicit buildCommand and startCommand when available
• ✅ Read port from package.json scripts or Dockerfile EXPOSE
• ✅ Detect static vs dynamic app type
• ✅ Warn when implicit defaults are used

Safety First

Claude SHOULD:

• ✅ Warn when secrets are referenced but not configured
• ✅ Warn when domain/SSL configuration is incomplete
• ✅ Verify health endpoint after deployment
• ✅ Provide rollback instructions on failure

Skill Structure

skill/
├── SKILL.md                    # This file - overview and principles
├── instructions/               # Step-by-step guides
│   ├── project-analysis.md    # How to inspect a project
│   ├── deploy-planning.md     # How to create deploy intent
│   ├── platform-selection.md   # How to choose Coolify vs Dokploy
│   └── post-deploy-verification.md
├── checklists/                 # Pre-deployment checks
│   ├── preflight-checklist.md
│   └── security-checklist.md
├── playbooks/                  # Deployment workflows by type
│   ├── node-app.md
│   ├── static-site.md
│   ├── dockerfile-app.md
│   └── compose-stack.md
└── examples/                   # Claude workflow examples
    └── node-to-deploy.yaml

Quick Reference

Claude Deployment Workflow

1. Receive deploy request
2. Read project files (package.json, Dockerfile, etc.)
3. Detect project type (static/source/dockerfile/compose)
4. Generate deploy.intent.yaml
5. Select platform (or ask user)
6. Create application via MCP
7. Set environment variables
8. Configure domain (if provided)
9. Trigger deployment
10. Poll for completion
11. Verify health endpoint
12. Report success/failure with logs

MCP Tool Sequences

Coolify:

list_projects → create_application_* → bulk_set_envs → assign_domain → trigger_deploy → get_deployment_status

Dokploy:

list_projects → create_application → bulk_set_envs → assign_domain → trigger_deploy → get_logs

See Also

• project-analysis.md
• deploy-planning.md
• platform-selection.md
• post-deploy-verification.md
• decision-tree.md
• failure-recovery.md