OpenCode Configuration

Personal OpenCode configuration for AI-assisted development on my work machine.

Overview

This repository contains my customized OpenCode setup featuring:

• Custom agents for specialized tasks (code review, planning)
• Skills for consistent AI behavior (Karpathy guidelines, security awareness)
• MCP servers for extended capabilities (database access)
• Model configuration optimized for different agent roles

Structure

.
├── AGENTS.md                       # Project knowledge base
├── LICENSE                         # MIT License
├── README.md                       # This file
├── opencode.jsonc                  # OpenCode workspace configuration
├── agents/
│   ├── build.md                # @build agent (autonomous implementation)
│   ├── explore.md              # @explore agent (enhanced)
│   ├── general.md              # @general agent (enhanced)
│   ├── librarian.md            # @librarian agent (external research)
│   ├── oracle.md               # @oracle agent (consultation)
│   └── plan.md                 # @plan agent (strategic planning)
└── skills/
    ├── ast-grep/               # Structural code search with AST patterns
    ├── coderabbit-code-review/ # CodeRabbit CLI code review
    ├── exa-search/             # Web search via Exa API
    ├── find-docs/              # Context7 documentation lookup
    ├── karpathy-guidelines/    # LLM coding best practices
    └── security-awareness/     # Security threat detection

Configuration

Models

Configured models follow OpenAI Codex best practices:

Agent	Model	Variant	Purpose
build	openai/gpt-5.3-codex	high	Complex implementation tasks
plan	openai/gpt-5.4	xhigh	Strategic planning
oracle	openai/gpt-5.4	high	Architecture consultation
general	openai/gpt-5.3-codex	medium	General-purpose work
explore	openai/gpt-5.1-codex-mini	—	Fast discovery
librarian	openai/gpt-5.1-codex-mini	—	External research

Agents

• @build: Autonomous implementation agent (primary)

  • Write, modify, and fix code with surgical precision
  • Delegates to @explore, @oracle, @librarian for parallel work
  • Verifies with linters, type checkers, tests

• @plan: Strategic planning agent (primary)

  • Analyzes complex tasks and coordinates work
  • Produces decision-complete execution plans
  • Interviews user, explores context before planning

• @oracle: Read-only high-IQ consultant

  • Architecture and complex decision consultation
  • Debugging after 2+ failed attempts, catches blind spots
  • Security/performance analysis with confidence levels
  • Multi-system tradeoff evaluation

• @librarian: External reference librarian

  • Official documentation lookup (Context7, API docs)
  • Production-quality OSS implementation examples
  • Library best practices and security guidance
  • Distinguishes documented facts from opinions

• @explore (built-in, enhanced): Contextual codebase exploration

  • Pattern discovery and conventions
  • Module relationship mapping
  • Pre-implementation research

• @general (built-in, enhanced): Fast execution agent for well-specified quick tasks

  • Quick task execution with minimal overhead
  • Following existing codebase patterns exactly
  • Surgical, minimal changes only
  • Evidence-based completion with verification

Skills

• ast-grep: Structural code search using Abstract Syntax Tree patterns

  • Find code patterns that text search cannot handle
  • Locate specific language constructs (e.g., "async functions without error handling")
  • Match code based on structure rather than just text
  • Requires ast-grep CLI: npm install -g @ast-grep/cli

• coderabbit-code-review: AI-powered code review using CodeRabbit CLI

  • Finds bugs, security issues, and quality risks
  • Groups findings by severity (Critical, Warning, Info)
  • Supports staged, committed, or all changes
  • Requires CodeRabbit CLI: npm install -g coderabbit

• exa-search: Web search via Exa API for research and current information

  • Semantic web search with filters
  • Fetch contents from URLs
  • Find similar links
  • Grounded Q&A with streaming support
  • Requires Exa CLI and EXA_API_KEY

• find-docs: Documentation lookup via Context7 for up-to-date library docs

  • Query current documentation for any library
  • Get code examples and usage patterns
  • Version-specific documentation support
  • Covers setup, migration guides, and API docs
  • Requires Context7 CLI: npm install -g ctx7

• karpathy-guidelines: Behavioral guidelines to reduce LLM coding mistakes

  • Think before coding
  • Simplicity first
  • Surgical changes
  • Goal-driven execution

• security-awareness: Security threat detection and credential protection

  • Phishing detection
  • Domain verification
  • Credential handling
  • Social engineering defense

MCP Servers

• dbeaver: Local DBeaver MCP server for database access (read-only)

Philosophy

This configuration follows these principles:

1. Minimal but complete: Start small, add only what's needed
2. Delegation-first: Encourage agents to use specialized subagents
3. Verification: Validate with tools, not just analysis
4. Security: Built-in awareness of common threats

Future Enhancements

Planned additions for the work machine setup:

• Additional subagents (oracle, librarian, explore, general)
• Custom build/plan prompts with delegation encouragement
• Additional MCP servers (context7, grep-app, exa)
• Code review via CodeRabbit CLI
• Documentation lookup via Context7 (find-docs skill)
• Web search via Exa (exa-search skill)
• AST-based code search (ast-grep skill)
• Workflow commands (/plan, /research, /consult)
• Build configuration package (template agents/commands/skills)

References

This configuration draws inspiration from:

• oh-my-opencode - Multi-agent orchestration concepts
• OpenAI Codex - Prompt structure and agent patterns
• Superpowers - Skill-based workflow enforcement
• oh-my-pi - System prompt design

License

MIT License - see LICENSE file for details.

---

Note: This is a personal configuration. Adapt to your needs. For more fun you might want to check my OmO config