OpenCode Vault is a curated vault of configurations that leverage OpenCode’s ability to execute skills, deploy persona-driven agents, and extend complex workflows. By defining precise behavioral boundaries within the .opencode directory, this project enables users to automate complex tasks (e.g. code reviews) without writing custom orchestration code.
Key Components
- Persona-Based Agent Configs (
.opencode/agents/): Pre-defined role definitions (Markdown + YAML frontmatter) that assign specific professional contexts to OpenCode (e.g., Security Auditor, Refactoring Specialist, Technical Copywriter) accessible via@agent-name. - Skill Manifests (
.opencode/skills/): A library of auto-discoverable, Claude-Code–compatible instructions (SKILL.md) providing atomic "how-to" units for specific tasks that agents call dynamically.
- OpenCode CLI or Desktop
- API Key to your preferred LLM provider (e.g. AWS Bedrock, Gemini, etc.)
- Ollama if running LLMs locally
-
Install OpenCode on your laptop (see OpenCode Documentation)
-
Make a copy of
./opencode.jsonc.templateas./opencode.jsoncto configure OpenCode (see OpenCode documentation)- Enter API keys for your LLM provider of choice
-
Install Ollama models locally
ollama pull <model-name>
where
<model-name>is the name of the model to be installed (e.g. "gemma4:e4b") -
Start OpenCode CLI or Desktop
Every skill is stored in its own folder under .opencode/skills and has the following structure:
your-skill-name/
├── SKILL.md # Required — the core instruction file
├── docs/ # Required - the documentation on how to use the skill
├── scripts/ # Optional — executable code
├── references/ # Optional — documents to loaded as needed
└── assets/ # Optional — templates, examples
Note that the folder name should be in kebab-case only (e.g. your-skill-name).