Supercharging Opencode with AWS Bedrock, Oh-My-Opencode, and Custom Skills

After a few months of experimenting with AI coding assistants, I’ve settled on a setup that significantly accelerates my development workflow. This post walks through configuring Opencode with AWS Bedrock, the oh-my-opencode plugin for multi-agent orchestration, and various plugins and skills that make the whole system more powerful.

Why This Setup?

#

A few reasons I landed on this particular combination:

1. Model flexibility: Opencode works with any LLM provider - Anthropic direct, AWS Bedrock, OpenAI, local models via Ollama. No vendor lock-in.
2. Cost management: AWS Bedrock with cross-region inference gives access to Claude Opus 4.5, Sonnet 4.5, and Haiku 4.5 with AWS billing - useful if you’re already in the AWS ecosystem.
3. Multi-agent orchestration: Oh-my-opencode adds specialized sub-agents that can work in parallel on different aspects of a task.
4. Extensibility: MCP servers, plugins, and skills let you customize the experience for your specific workflows.

The Architecture

#

The setup consists of several layers:

Prerequisites

#

Before starting, you’ll need:

• Node.js v18 or later
• An AWS account with Bedrock access enabled
• AWS CLI configured with credentials (aws configure)
• Basic comfort with the terminal

Step 1: Install Opencode

#

# Install globally via npm
npm install -g @opencode-ai/cli

# Or use npx to run directly
npx opencode

Verify the installation:

opencode --version

Step 2: Configure AWS Bedrock

#

Opencode supports AWS Bedrock out of the box. The model format is amazon-bedrock/<model-id>.

For cross-region inference (recommended for better availability), use the global. prefix:

amazon-bedrock/global.anthropic.claude-opus-4-5-20251101-v1:0
amazon-bedrock/global.anthropic.claude-sonnet-4-5-20250929-v1:0
amazon-bedrock/global.anthropic.claude-haiku-4-5-20251001-v1:0

Make sure your AWS credentials are configured:

# Check your current identity
aws sts get-caller-identity

# If not configured, run:
aws configure

You’ll need appropriate IAM permissions for Bedrock. At minimum:

• bedrock:InvokeModel
• bedrock:InvokeModelWithResponseStream

Step 3: Install Oh-My-Opencode

#

Oh-my-opencode is the plugin that transforms Opencode into a multi-agent system. It adds specialized agents that can work in parallel on different aspects of your tasks.

# Add to your opencode config
# In ~/.config/opencode/opencode.json, add to the plugin array:

Edit ~/.config/opencode/opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": [
    "oh-my-opencode@latest"
  ]
}

Configure Agent Models

#

Create ~/.config/opencode/oh-my-opencode.json to specify which models each agent uses:

{
  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
  "agents": {
    "sisyphus": {
      "model": "amazon-bedrock/global.anthropic.claude-opus-4-5-20251101-v1:0"
    },
    "oracle": {
      "model": "amazon-bedrock/global.anthropic.claude-sonnet-4-5-20250929-v1:0"
    },
    "librarian": {
      "model": "amazon-bedrock/global.anthropic.claude-sonnet-4-5-20250929-v1:0"
    },
    "explore": {
      "model": "amazon-bedrock/global.anthropic.claude-haiku-4-5-20251001-v1:0"
    }
  },
  "categories": {
    "visual-engineering": {
      "model": "amazon-bedrock/global.anthropic.claude-sonnet-4-5-20250929-v1:0"
    },
    "ultrabrain": {
      "model": "amazon-bedrock/global.anthropic.claude-opus-4-5-20251101-v1:0"
    },
    "quick": {
      "model": "amazon-bedrock/global.anthropic.claude-haiku-4-5-20251001-v1:0"
    }
  }
}

The Agent Hierarchy

#

Oh-my-opencode provides several specialized agents:

The key insight is model tiering: use expensive models (Opus) for complex reasoning, mid-tier (Sonnet) for most tasks, and cheap models (Haiku) for exploration and simple operations.

Step 4: Add Useful Plugins

#

Plugins extend Opencode’s capabilities at the system level. Here are some I find useful:

{
  "plugin": [
    "oh-my-opencode@latest",
    "@mathew-cf/opencode-terminal-notifier",
    "envsitter-guard@latest",
    "@tarquinen/opencode-smart-title",
    "opencode-wakatime"
  ]
}

Step 5: Configure MCP Servers

#

MCP (Model Context Protocol) servers give Opencode access to external tools and data sources. Add them to your opencode.json:

{
  "mcp": {
    "awslabs.aws-documentation-mcp-server": {
      "type": "local",
      "command": ["uvx", "awslabs.aws-documentation-mcp-server@latest"],
      "environment": {
        "FASTMCP_LOG_LEVEL": "ERROR"
      },
      "enabled": true,
      "timeout": 120000
    },
    "awslabs.cdk-mcp-server": {
      "type": "local",
      "command": ["uvx", "awslabs.cdk-mcp-server@latest"],
      "enabled": true
    },
    "smart-connections": {
      "type": "local",
      "command": ["node", "/path/to/smart-connections-mcp/dist/index.js"],
      "environment": {
        "SMART_VAULT_PATH": "/path/to/obsidian/vault"
      },
      "enabled": true
    }
  }
}

Some MCP servers I use regularly:

Step 6: Install Skills

#

Skills are reusable knowledge modules that agents can load on-demand. They’re markdown files that give agents specialized knowledge about specific topics.

# Skills are stored in ~/.config/opencode/skills/
# Each skill is a directory with a SKILL.md file

# Install a skill from the ecosystem
npx skills add jackspace/claudeskillz@hugo

# Or create your own
mkdir -p ~/.config/opencode/skills/my-skill
# Add SKILL.md with your specialized knowledge

Skills I currently have installed:

• hugo - Static site generation with Hugo
• git-commit - Conventional commit message formatting
• manimce-best-practices - Manim animation library patterns
• obsidian - Obsidian vault interaction

Step 7: Create an AGENTS.md

#

The AGENTS.md file in ~/.config/opencode/ provides global instructions that all agents follow. This is where you codify your engineering principles:

# Engineering Principles

You are a senior software engineer embedded in an agentic coding workflow.

## Core Behaviors

### Assumption Surfacing (Critical)
Before implementing anything non-trivial, explicitly state your assumptions:
- Never silently fill in ambiguous requirements
- Surface uncertainty early

### Push Back When Warranted
You are not a yes-machine. When the human's approach has clear problems:
- Point out the issue directly
- Propose an alternative
- Accept their decision if they override

### Simplicity Enforcement
Prefer the boring, obvious solution. Cleverness is expensive.

This ensures consistent behavior across all agents and sessions.

Practical Workflows

#

Parallel Exploration

#

When tackling an unfamiliar codebase, fire multiple explore agents in parallel:

> Search for authentication patterns in this repo
> Also look for how error handling is done
> And find the database access patterns

The explore agent (running Haiku) quickly surfaces patterns while you focus on the main task.

Delegating to Specialists

#

For complex tasks, the main agent can delegate to specialists:

> Implement a new REST endpoint for user preferences

Behind the scenes, Sisyphus might:

1. Fire explore to understand existing endpoint patterns
2. Consult oracle for architecture decisions
3. Delegate implementation to a sub-agent
4. Run verification

Research-Heavy Tasks

#

For tasks requiring external knowledge:

> How should I implement DynamoDB single-table design for this use case?

The librarian agent queries documentation, finds examples on GitHub, and synthesizes recommendations - all without you leaving the terminal.

Cost Considerations

#

With AWS Bedrock, you pay per token. Model tiering helps manage costs:

My configuration uses Haiku for explore and quick categories, Sonnet for most work, and reserves Opus for ultrabrain and main orchestration.

Troubleshooting

#

“Model not found” errors

#

Ensure you’ve enabled the specific Claude models in your AWS Bedrock console. Cross-region inference models require explicit enablement.

Slow responses

#

Check your AWS region. Cross-region inference (global. prefix) routes to the nearest available region but can add latency. Consider using region-specific model IDs if latency is critical.

Plugin installation issues

#

# Clear plugin cache
rm -rf ~/.config/opencode/node_modules
rm ~/.config/opencode/package-lock.json

# Reinstall
opencode

MCP server timeouts

#

Increase the timeout in your config:

{
  "mcp": {
    "server-name": {
      "timeout": 180000
    }
  }
}

Resources

#

• Opencode Documentation - Official docs
• oh-my-opencode GitHub - Multi-agent plugin
• AWS Bedrock Claude Models - Model availability
• MCP Specification - Model Context Protocol

This setup has become central to my daily development workflow. Curious to hear how others are configuring their agentic development environments - what models, plugins, or workflows have you found effective?