OpenCode Best Practices

The most comprehensive collection of agents, skills, and tooling for OpenCode -- the open-source AI coding agent.

108 agents | 15 skills | 18 curated MCP servers | Live MCP registry search | Smart model detection

Inspired by VoltAgent/awesome-claude-code-subagents, fully adapted for OpenCode with permissions, skills, and markdown agent format.

Quickstart

npx @weisser-dev/awesome-opencode

One command. It analyzes your project, walks you through agent/skill/model/MCP setup interactively, generates everything, and launches OpenCode.

What it Does

  Awesome OpenCode
  108 agents, 15 skills, smart model config
  https://github.com/weisser-dev/awesome-opencode

? How would you like to set up your project?
❯ Auto-detect languages (recommended for existing projects)
  Select languages manually (recommended for fresh projects)

✔ Project analyzed

  Auto-detected:
    - TypeScript
    - Python

? Select project languages (2 detected / 28 total, scroll with arrows):
  ── Auto-Detected (2) ──
    ◉ JavaScript / TypeScript
    ◉ Python
  ── Other (26) ──
    ◯ Java
    ◯ Go
    ◯ Rust
    ...

? Select agents (12 recommended / 108 total, scroll with arrows):
  ── Recommended for your project (12/108) ──
    ◉ code-reviewer - Code quality review (default)
    ◉ test-writer - Test generation (default)
    ◉ devops-engineer - CI/CD pipelines (default)
    ◉ typescript-pro - TypeScript specialist (JavaScript / TypeScript)
    ◉ python-pro - Python ecosystem master (Python)
    ◉ fastapi-developer - FastAPI async APIs (Python)
    ...
  ── Core Development (3) ──
    ◯ backend-developer
    ...
  (108 agents across 10 categories)

? Select skills (5 recommended / 15 total, scroll with arrows):
    ◉ git-release - Release notes and version bumps
    ◉ ci-pipeline - CI pipeline configuration
    ◉ dependency-audit - Audit dependencies for vulnerabilities
    ◉ test-patterns - Test generation following project conventions
    ◉ changelog-generate - Changelog generation from commit history
    ◯ pr-review
    ...

  Models found in your config:

    my-provider/eu.anthropic.claude-sonnet-4-6
      -> claude-sonnet-4 (anthropic) | Strong | Cost: $$$ | Coding: 90/100
    my-provider/eu.anthropic.claude-haiku-4-5-20251001-v1:0
      -> claude-haiku-4 (anthropic) | Fast   | Cost: $$  | Coding: 75/100
    my-provider/eu.anthropic.claude-opus-4-6-v1
      -> claude-opus-4 (anthropic) | Frontier | Cost: $$$$ | Coding: 95/100

? Model strategy:
❯ Auto-optimize YOUR models (Build: claude-opus-4, Plan/Explore: claude-haiku-4)
  Cost Optimized (Haiku for Plan/Explore, Sonnet for Build)
  Quality Focused (Opus for Build, Sonnet for others)
  OpenAI Stack
  Mixed
  Keep existing

? Select MCP servers (14 available for your languages, scroll with arrows):
  ── Universal ──
    ◉ context7 - Library documentation search
    ◉ git - Read, search, and manipulate Git repos
    ◯ memory - Persistent knowledge graph (official)
    ...
  ── Database ──
    ◯ postgres - PostgreSQL read-only access (official)
    ...

? Search mcp.so for additional MCP servers? Yes
? Search mcp.so: playwright
  Found 45 server(s) on mcp.so:
? Select servers to add (45 found, scroll with arrows):
    ◯ Playwright Mcp (microsoft)
    ...

? Apply recommended step limits per agent? (controls context/cost) Yes
  Code-writing agents: unlimited | Review agents: 10-15 | Fast agents: 5-10

✔ No AGENTS.md found. Generate one with project-specific rules? Yes

  Generated:
    .opencode/agents/ (5 agents)
    .opencode/skills/ (3 skills)
    opencode.json (updated)
    .opencode/advanced.json (state)
    AGENTS.md (generated)

  Setup complete!

? Run OpenCode in a sandbox? (Docker, recommended for enterprise) No
? Start OpenCode? Yes

  Starting OpenCode...

On re-run, it remembers your setup:

  Awesome OpenCode
  108 agents, 15 skills, smart model config
  https://github.com/weisser-dev/awesome-opencode

  Already configured!
  Last setup: 2026-03-27
  Languages:  node, python
  Agents:     code-reviewer, test-writer, typescript-pro, python-pro
  Skills:     git-release, ci-pipeline, dependency-audit
  Models:     auto

? What would you like to do?
❯ Start OpenCode
  Start OpenCode (Sandboxed via Docker)
  Reconfigure (run setup again)
  Configure agents
  Configure skills
  Configure models
  Configure MCP servers
  Exit

Features

108 Agents in 10 Categories

15 Skills

Smart Model Detection

The CLI recognizes models from any provider -- including custom Bedrock, Azure, or self-hosted endpoints:

abcd/eu.anthropic.claude-opus-4-6-v1     -> claude-opus-4   | Frontier | $$$$  | Coding: 95/100
abcd/eu.anthropic.claude-sonnet-4-6      -> claude-sonnet-4 | Strong   | $$$   | Coding: 90/100
abcd/eu.anthropic.claude-haiku-4-5       -> claude-haiku-4  | Fast     | $$    | Coding: 75/100

Supports 26+ model fingerprints across Anthropic, OpenAI, Google, DeepSeek, Meta, and Mistral.

Agent-tier mapping assigns the right model quality per agent type:

• Frontier (Opus, GPT-5.2, o3): Code-writing agents, language specialists
• Strong (Sonnet, GPT-5.1): Code review, architecture, security
• Fast (Haiku, Flash, mini): Exploration, docs, context management

18 Curated MCP Servers + Live Registry Search

MCP servers are filtered by your selected languages -- Terraform projects see AWS/K8s, Node projects see Vercel/Figma.

Plus: live search of the official MCP Registry (registry.modelcontextprotocol.io) to find and install any registered server.

Documentation

Installation

Via npx (recommended)

npx @weisser-dev/awesome-opencode

Global install

npm install -g @weisser-dev/awesome-opencode
awesome-opencode

CLI Subcommands

awesome-opencode                      # Interactive setup or re-run menu
awesome-opencode configure            # Reconfigure everything
awesome-opencode configure agents     # Add/remove agents
awesome-opencode configure skills     # Add/remove skills
awesome-opencode configure models     # Change model strategy
awesome-opencode configure mcp        # Add/remove MCP servers
awesome-opencode --help               # Show help

Flags

awesome-opencode --config <path>      # Use external opencode.json
awesome-opencode --crt <path>         # Custom CA certificate (.crt/.pem)
awesome-opencode --skipSSL            # Disable TLS verification

Flags can be combined with subcommands and with each other:

# Corporate setup: custom cert + external config
awesome-opencode --crt /etc/ssl/corporate-ca.crt --config ~/shared/opencode.json

# Behind proxy with self-signed cert
awesome-opencode --skipSSL configure mcp

# All flags are forwarded into Docker sandbox automatically
awesome-opencode --crt /certs/ca.pem --skipSSL

Sandboxed Mode (Docker)

Run OpenCode in an isolated Docker container where only the current project is accessible:

? Select dev environment (Docker image):
  ── Recommended for your project ──
    ❯ Java 21 + Maven (Spring Boot, enterprise)
      Java 21 + Gradle
  ── Other environments ──
      Node.js 22 (JavaScript, TypeScript)
      Python 3.13 (pip)
      ...
  ── General purpose ──
      Ubuntu 24.04 (apt available — install anything)
      Generic Node.js 22 (works for any project)

? Which LLM provider are you using?
❯ AWS Bedrock

? Configure corporate proxy? No
? Configure artifact registry (Nexus, JFrog, PyPI mirror)? Yes
? Select artifact registries:
    ◉ Nexus Repository (Maven)
? MAVEN_MIRROR_URL (required): https://nexus.example.com/repository/maven-public/

  Docker command:
  docker run -it --rm \
    -v "$PWD":"$PWD" \
    -w "$PWD" \
    -e AWS_BEARER_TOKEN_BEDROCK \
    -e AWS_REGION="eu-central-1" \
    -e MAVEN_MIRROR_URL="https://nexus.example.com/repository/maven-public/" \
    maven:3-eclipse-temurin-21 \
    bash -c "mkdir -p ~/.m2 && echo '<settings>...</settings>' > ~/.m2/settings.xml && apt-get install -y nodejs npm && npm i -g opencode-ai && opencode"

? Run this Docker command now? Yes

20 supported dev environments — all official images:

Supports 7 LLM providers: Anthropic, OpenAI, AWS Bedrock, Azure OpenAI, Google AI, OpenRouter, Custom. Supports corporate proxy + artifact registries: Nexus (npm + Maven), JFrog (npm + Maven), PyPI mirror.

Manual setup (without CLI)

# 1. Copy agents
mkdir -p .opencode/agents
cp templates/agents/code-reviewer.md .opencode/agents/
cp templates/agents/typescript-pro.md .opencode/agents/

# 2. Copy skills
mkdir -p .opencode/skills/git-release
cp templates/skills/git-release/SKILL.md .opencode/skills/git-release/

# 3. Copy config
cp templates/configs/cost-optimized.json opencode.json

# 4. Generate AGENTS.md
opencode    # then /init in the TUI

Development

git clone https://github.com/weisser-dev/awesome-opencode
cd awesome-opencode/cli-tool
npm install
npm link

# Test in any project:
cd ~/my-project
awesome-opencode

# Run tests:
npm test

# Unlink when done:
npm unlink -g @weisser-dev/awesome-opencode

Publishing to npm

cd cli-tool
npm run prepublishOnly   # sync templates from repo root
npm pack --dry-run       # verify contents
npm publish              # publish (requires npm login + org access)

Project Structure

awesome-opencode/
  README.md                    # This file
  CHANGELOG.md                 # Release history
  AGENTS.md                    # Project rules for OpenCode
  LICENSE                      # MIT
  docs/                        # Comprehensive documentation (7 files)
  templates/
    agents/                    # 108 agent definitions (.md)
    skills/                    # 15 skill definitions (SKILL.md)
    configs/                   # 5 opencode.json presets
  cli-tool/                    # @weisser-dev/awesome-opencode
    src/
      index.js                 # Entry point and flow control
      setup.js                 # Detection, prompts, generation (~1800 lines)
    templates/                 # Bundled copy for npm package
    scripts/
      sync-templates.js        # Pre-publish template sync
    package.json

Inspiration & Credits

This project builds on the work of several open-source communities:

• VoltAgent/awesome-claude-code-subagents -- The agent collection (127+ subagents) served as the primary inspiration. All agents were adapted for OpenCode with permissions, skill integration, and markdown format.
• darrenhinde/OpenAgentsControl -- Agent configuration patterns and workflows.
• OpenCode Official Docs -- Agents, skills, MCP, permissions, and configuration reference.
• pricepertoken.com -- Model benchmark rankings and pricing data for the model intelligence feature.
• models.dev -- Open-source database of AI models (by anomalyco). Model fingerprints sourced from here.
• registry.modelcontextprotocol.io -- Official MCP server registry API for live search.

Also by weisser-dev

• agentic-ai.weisser.dev -- Free, ad-free self-learning platform for Agentic AI. Want to learn more about agents, MCP, and AI-assisted development? Start here.
• opencode-remote-telegram -- Control OpenCode remotely via Telegram. Run sessions from your phone.

Resources

• OpenCode Docs
• OpenCode GitHub
• OpenCode Discord
• OpenCode Ecosystem
• MCP Server Directory
• Official MCP Registry
• pricepertoken.com -- LLM pricing and benchmarks
• VoltAgent Subagents

License

MIT

Contributing

Pull requests welcome! Especially for:

• New agent templates (add to templates/agents/ and AVAILABLE_AGENTS in setup.js)
• New skill definitions (add to templates/skills/<name>/SKILL.md and AVAILABLE_SKILLS)
• Framework-specific config presets
• Model fingerprint additions (new providers/models)
• MCP server recommendations
• Documentation improvements
• Translations