Getting Started
Requirements
- Bun runtime (>=1.0.0)
- OpenCode CLI
- AI agent supporting Agent Skills (OpenCodeAI, Claude Code, etc.)
Installation
Detailed installation instructions are available in the Installation guide.
Quick start:
bun run installThis runs an interactive installer that lets you choose between:
- Local -
.opencode/(current project only) - Global -
~/.config/opencode/(all projects) - Custom - Specify any installation path
What Gets Installed
Every request goes through a structured execution flow:
- Classify Intent — What do you want?
- Load Context — Project rules for this task
- Load Skill — Appropriate specialist
- Execute — Do the work (follows Plan when applicable)
- Reflect — Revisit, rethink, re-evaluate
Plan Framework (when structured development needed):
- PLAN — Define objectives and acceptance criteria (Given/When/Then)
- APPLY — Execute with verification steps
- UNIFY — Close loop, reconcile plan vs actual, update state
CARL Quality Gates (active throughout):
- Dynamic rule loading based on domain
- Priority-based enforcement (Critical > High > Medium)
- Blocks on critical violations
Structure at the start, freedom at the end.
Running Scripts
AI Agent
The AI agent can run Tachikoma scripts directly as tools:
tachikoma.where
tachikoma.edit-format-selector recommendManual
Use bun run for direct script execution:
bun run .opencode/plugins/tachikoma/where.ts
bun run .opencode/plugins/tachikoma/edit-format-selector.ts recommendUnderstanding Plan Methodology
Plan (Plan-Apply-Unify Loop) — Structured development framework for AI-assisted workflows.
- PLAN — Define objectives and acceptance criteria (Given/When/Then format)
- APPLY — Execute tasks sequentially with verification
- UNIFY — Close loop, reconcile plan vs actual, update state
Never skip UNIFY — this is the heartbeat that prevents drift.
Quality over speed-for-speed's-sake. In-session context over subagent sprawl.
Understanding CARL
CARL (Context Augmentation & Reinforcement Layer) — Dynamic rule loading system.
CARL loads quality rules just-in-time based on context:
- Context Detection — Detects active domains (Plan, Development, Projects)
- Rule Loading — Loads relevant rules dynamically
- Priority Enforcement — Critical blocks, high warns, medium notes
Three Domains:
- Plan Domain — Loop enforcement, boundary protection
- Development Domain — Code quality, error handling, testing
- Projects Domain — Documentation, version handling
Next Steps
- Concepts Overview — Understand the architecture
- Capabilities Index — Explore all features
- PAUL Methodology — Learn the framework
- Intent Routing — How requests are classified
- Internals — Database schema and internals