advanced⏱60 minLección 6 de 10

OpenCode as Development Tool

Learn to set up, configure, and use OpenCode for AI-assisted software development with custom agents, skills, and pipelines.

OpenCode as Development Tool

What is OpenCode?

OpenCode is an open-source CLI framework for AI-assisted software engineering. It provides a structured environment where AI agents interact with your codebase through a permission-controlled tool system, reusable skills, and the Model Context Protocol (MCP) for external integrations.

💡 Tip: Use Ctrl + Mouse Wheel to zoom, or click the fullscreen icon

ℹ️Note

OpenCode can be installed globally via npm or pip and configured per-project using opencode.json at the project root or .opencode/config.json. The CLI command is opencode (or oc for short).

Installation and Setup

bash

npm install -g opencode-cli
pip install opencode-cli
opencode --version
opencode init
opencode

json

{
  "name": "my-ai-powered-project",
  "version": "1.0.0",
  "scripts": {
    "dev": "opencode --agent developer",
    "review": "opencode --agent reviewer --task 'Review all new changes'",
    "deploy": "opencode --agent devops --task 'Deploy to staging'"
  },
  "opencode": {
    "configPath": ".opencode/config.json"
  }
}

💡Tip

Use opencode init to generate a starter opencode.json with sensible defaults. It prompts for your preferred model provider and creates the basic agent structure.

Core Configuration

json

{
  "opencode": "1.0",
  "agents": {
    "default": {
      "model": "gpt-4o",
      "description": "General-purpose coding assistant",
      "prompt": "You are an expert software engineer. Always write clean, well-tested code."
    },
    "reviewer": {
      "model": "claude-sonnet-4-20250514",
      "description": "Code review and security audit specialist",
      "prompt": "You are a senior code reviewer. Focus on security, performance, and maintainability.",
      "constraints": {
        "allowedTools": ["read", "grep", "glob"],
        "deniedTools": ["write", "edit", "bash"]
      }
    },
    "architect": {
      "model": "gpt-4o",
      "description": "Software architecture and design specialist",
      "prompt": "You are a software architect. Focus on system design, patterns, and scalability.",
      "constraints": { "temperature": 0.4 }
    }
  },
  "permissions": [
    {
      "tool": "bash",
      "allow": ["npm *", "git *", "pip *", "python *", "pytest *"],
      "deny": ["rm -rf /", "sudo *", "chmod *"]
    },
    {
      "tool": "write",
      "allow": ["src/**", "tests/**", "docs/**"],
      "deny": [".env", "secrets/**", "node_modules/**"]
    },
    {
      "tool": "edit",
      "allow": ["src/**", "tests/**"],
      "deny": ["*.lock", "package-lock.json"]
    }
  ],
  "skills": {
    "code-reviewer": {
      "manifest": "skills/code-reviewer/skill.yaml",
      "autoLoad": true,
      "matchPattern": "review|audit|inspect|check"
    },
    "test-generator": {
      "manifest": "skills/test-generator/skill.yaml",
      "autoLoad": true,
      "matchPattern": "test|spec|coverage"
    }
  },
  "mcpServers": {
    "github": {
      "command": "node",
      "args": ["mcp-github-server.js"],
      "env": { "GITHUB_TOKEN": "${GITHUB_TOKEN}" }
    },
    "filesystem": {
      "command": "node",
      "args": ["mcp-server-fs.js", "/home/user/projects"]
    }
  },
  "agentRouting": {
    "mode": "auto",
    "defaultAgent": "default",
    "rules": [
      { "pattern": "security|vulnerability|CVE|OWASP", "agent": "reviewer" },
      { "pattern": "architecture|design|pattern|scalability", "agent": "architect" }
    ]
  }
}

Agent Routing Workflow

💡 Tip: Use Ctrl + Mouse Wheel to zoom, or click the fullscreen icon

ℹ️Note

Routing happens at the beginning of each request. Override routing with --agent <name> to force a specific agent, or --skill <name> to load a specific skill.

Working with Skills

yaml

# skills/react-component/skill.yaml
name: react-component
description: Generate React components with TypeScript, tests, and stories
version: "1.0.0"
instructions: |
  When asked to create a React component:
  1. Determine the component name and props interface
  2. Create src/components/{Name}/{Name}.tsx
  3. Create src/components/{Name}/{Name}.test.tsx
  4. Create src/components/{Name}/{Name}.stories.tsx
  5. Export from src/components/index.ts

  Follow these rules:
  - Use TypeScript with strict types
  - Include loading, empty, error, and edge case states
  - Follow existing component patterns
  - Add JSDoc comments for public APIs
tools:
  required:
    - read
    - write
    - glob
  optional:
    - bash
resources:
  - path: templates/component.tsx.hbs
  - path: templates/test.tsx.hbs
  - path: templates/story.tsx.hbs
autoload:
  enabled: true
  matchPattern: "create component|new component|react component"

bash

opencode --skill react-component "Create a Button component with variants"
opencode --list-skills
opencode --skill-info react-component

⚠️Warning

Skills load automatically based on match patterns. If a pattern is too broad, the skill may activate in unintended contexts. Keep match patterns specific.

Permission System in Practice

json

{
  "permissions": [
    {
      "tool": "bash",
      "description": "Shell command execution",
      "allow": ["npm *", "git *", "pip *", "python *", "pytest *"],
      "deny": ["rm -rf /", "sudo *", "chmod 777 *"]
    },
    {
      "tool": "write",
      "allow": ["src/**/*.ts", "src/**/*.py", "tests/**/*.py", "docs/**/*.md"],
      "deny": [".env", "secrets/**", "node_modules/**", "*.key", "*.pem"]
    }
  ],
  "approvalGates": [
    {
      "description": "Production deployment gate",
      "condition": "command matches 'kubectl apply -f k8s/production/*'",
      "requireApprovalFrom": "user"
    }
  ]
}

💡Tip

Start restrictive and expand gradually. Begin by denying all write/edit operations, then add write capabilities for specific file patterns as you validate agent behavior.

MCP Server Integration

javascript

// mcp-github-server.js
import { Server } from "@modelcontextprotocol/sdk/server/index.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import { Octokit } from "@octokit/rest";

const server = new Server({
  name: "github-mcp-server",
  version: "1.0.0",
}, { capabilities: { tools: {} } });

const octokit = new Octokit({ auth: process.env.GITHUB_TOKEN });

server.setRequestHandler("tools/list", async () => ({
  tools: [
    {
      name: "create_issue",
      description: "Create a GitHub issue",
      inputSchema: {
        type: "object",
        properties: {
          title: { type: "string" },
          body: { type: "string" },
          labels: { type: "array", items: { type: "string" } }
        },
        required: ["title"]
      }
    },
    {
      name: "search_code",
      description: "Search code in repository",
      inputSchema: {
        type: "object",
        properties: {
          query: { type: "string" },
          repo: { type: "string" }
        },
        required: ["query"]
      }
    }
  ]
}));

server.setRequestHandler("tools/call", async (request) => {
  const { name, arguments: args } = request.params;
  switch (name) {
    case "create_issue": {
      const { data: issue } = await octokit.issues.create({
        owner: "my-org", repo: args.repo || "default",
        title: args.title, body: args.body, labels: args.labels
      });
      return { content: [{ type: "text", text: JSON.stringify(issue) }] };
    }
    case "search_code": {
      const { data: search } = await octokit.search.code({ q: args.query });
      return { content: [{ type: "text", text: JSON.stringify(search.items.slice(0, 5)) }] };
    }
    default:
      throw new Error("Unknown tool: " + name);
  }
});

const transport = new StdioServerTransport();
await server.connect(transport);

json

{
  "mcpServers": {
    "github": {
      "command": "node",
      "args": ["mcp-github-server.js"],
      "env": { "GITHUB_TOKEN": "${GITHUB_TOKEN}" }
    },
    "postgres": {
      "command": "python",
      "args": ["mcp-postgres-server.py"],
      "env": { "DATABASE_URL": "${DATABASE_URL}" }
    }
  }
}

CLI Usage Patterns

bash

# Interactive mode
opencode

# Single command mode
opencode "Refactor the payment module to use async/await"

# Pipe input
cat requirements.txt | opencode "Generate a setup.py from this"

# File as context
opencode --file src/main.py "Review this file for bugs"

# Multi-file context
opencode --files src/*.py "Find unused imports"

# Agent selection
opencode --agent architect "Design database schema for multi-tenant app"

# Skill execution
opencode --skill code-reviewer "Review latest git diff"

# Debug mode
opencode --debug "Why is the build failing?"

# Output formats
opencode --format json "List all TODO comments" > todos.json
opencode --format markdown "Generate API documentation" > api-docs.md

✅Success

OpenCode transforms your development workflow by providing a structured, secure, and extensible platform for AI-assisted software engineering.

Best Practices for OpenCode

yaml

best_practices:
  configuration:
    - Use .opencode/ directory instead of root opencode.json
    - Version control config but use env vars for secrets
    - Define specific agents for specific tasks
    - Use matchPattern on skills for auto-activation
    - Start restrictive and expand gradually

  agent_design:
    - Give agents clear, specific descriptions for accurate routing
    - Use temperature 0.2-0.4 for deterministic code tasks
    - Use temperature 0.7-0.9 for creative tasks
    - Define constraints properly
    - Use subagents for specialized subtasks

  skill_development:
    - Keep skills focused on single responsibilities
    - Include edge cases and error handling in instructions
    - Version skills with semantic versioning
    - Test skills independently with mock agents
    - Bundle templates and reference files as resources

  security:
    - Never store secrets in opencode.json
    - Use approval gates for destructive operations
    - Deny write access to configuration and lock files
    - Audit permission rules regularly

Practice Exercises

Practice Question

What are the two equivalent locations for OpenCode configuration?

Practice Question

Which CLI flag forces a specific agent to handle a request?

Practice Question

What is the purpose of approvalGates in OpenCode permissions?

Practice Question

How does auto-loading work for skills in OpenCode?

Practice Question

What happens when both allowedTools and deniedTools are specified for an agent?

Practice Question

Which command initializes OpenCode in a new project?

Practice Question

What is the recommended approach for handling secrets in OpenCode configuration?

Practice Question

What protocol do MCP servers use to communicate with OpenCode?

✅**Key Takeaways**

OpenCode is configured via opencode.json at project root or .opencode/config.json
Agent routing uses description matching and rule patterns to dispatch tasks
Skills auto-load when user requests match their matchPattern
Permissions allow/deny specific tool operations with path patterns
Approval gates add human-in-the-loop for high-risk operations
MCP servers extend capabilities via JSON-RPC protocol
Use --agent to force routing, --skill for explicit skill loading
Environment variables keep secrets out of version-controlled config
Start with restrictive permissions and expand gradually
OpenCode transforms AI-assisted development with structured, secure agent workflows

Progreso60%

Agent Pipelines

Building Custom Agent Skills

Contenido del Curso

1. Introduction to AI Agents 2. Agent Architecture 3. Skills & Capabilities 4. Context Management 5. Agent Pipelines 6. OpenCode as Development Tool 7. Building Custom Agent Skills 8. Prompt Engineering for Agents 9. Best Practices & Patterns 10. Real-World Agent Projects