Opencode Configuration

This repository contains the personalized configuration and custom extensions for the Opencode AI development environment. It defines custom agents, commands, skills, and behavior rules to tailor the AI's capabilities for automated development, QA testing, and project planning.

Directory Structure

~/.config/opencode/
├── AGENTS.md           # Global instructions, memory management, and agent routing rules
├── opencode.json       # Main configuration file (plugins, MCP, permissions)
├── agents/             # Custom subagent definitions
│   ├── docs-writer.md       # Agent for technical documentation
│   └── playwright-tester.md # Agent for QA and Playwright automation
├── commands/           # Custom executable commands
│   ├── execute.md           # Task execution and verification workflow
│   └── fleshout.md          # Roadmap planning and task breakdown workflow
├── skills/             # Custom AI skills
│   └── playwright-cli/      # Skill for browser automation and test generation
└── package.json        # Plugin dependencies (e.g., @opencode-ai/plugin)

Key Features

1. Global Agent Instructions (AGENTS.md)

Enforces standard operating procedures across all sessions:

• Verifies clean Git status before initiating workflows.
• Routes specific tasks to specialized subagents (@docs-writer for docs, @playwright-tester for UI testing).
• Integrates continuous memory management (mnemosyne) using memory_recall and memory_store to maintain cross-session context and personal preferences.

2. Custom Agents

• @docs-writer: A specialized subagent focused solely on creating and maintaining technical documentation, READMEs, and guides. It is configured to safely write .md files even from read-only contexts.
• @playwright-tester: An expert QA Automation subagent equipped with the playwright-cli skill. It analyzes UI code, writes semantic tests, runs verifications, and manages end-to-end browser automation.

3. Custom Commands

• execute: Driven by the build agent, this command sequentially processes task specifications or roadmap items, verifying acceptance criteria, and explicitly enforcing linting, testing, and UI verification.
• fleshout: Driven by the docs-writer agent, this command breaks down high-level project roadmap items into detailed, actionable tasks with clear acceptance criteria and dependencies.

4. Custom Skills

• playwright-cli: A deeply integrated skill allowing agents to automate browser interactions without writing boilerplate code. It supports navigation, form filling, snapshot analysis, session management, mock networking, DevTools tracing, and automatic test generation.

5. Security & Permissions (opencode.json)

Overrides default agent permissions to maintain a secure automated environment:

• Git Protections: Explicitly denies automated git commit and git push across all agents to prevent unapproved repository modifications.
• Bash Execution: Extends timeout limits for long-running build operations while defaulting most bash commands to require user approval.

6. Integrations & Plugins

Configures Model Context Protocol (MCP) integrations:

• Context7: Local MCP for code understanding.
• Tavily: Remote MCP for web search and research.
• Also integrates local plugins (e.g. mnemosyne memory plugin) and the opencode-notifier for state notifications.

Installation & Setup

If you are cloning this configuration to a new machine:

1. Clone or copy this repository to ~/.config/opencode.
2. Install the necessary node dependencies using Bun or npm:

   bun install
   # or
   npm install

3. Ensure the required CLI tools (like playwright-cli, npx, etc.) are available in your global environment.
4. Restart your Opencode session to apply the new configuration.

Usage Guide

• Triggering Commands: You can invoke custom commands directly in your Opencode prompt (e.g., /execute or /fleshout).
• Invoking Agents: Mention @docs-writer or @playwright-tester in your prompt to delegate tasks specifically to them.
• Memory Context: The system will automatically recall relevant global memories at the start of a session. You can explicitly ask the AI to "remember this" to store custom preferences.
• Git Workflow: Since automated commits are blocked by default, the AI will stage files and present a git status summary. You must manually execute or approve the final commit.