OpenCode Configuration

Multi-agent development system for OpenCode with 13 specialized agents, 13 skills, and 12 commands organized around research → plan → review → implement → verify pipelines.

---

Quick Start

# 1. Clone or copy this config to ~/.config/opencode
# 2. Install dependencies
pnpm install

# 3. Verify
pnpm check

Requires: OpenCode, Node.js, pnpm, Git. Recommended: GitHub CLI (gh), ripgrep (rg).

---

Structure

~/.config/opencode/
├── opencode.json          # Agents, permissions, models, plugin
├── agents/                # 13 agent definitions (generated from src/)
├── commands/              # 12 commands (lanes, spec-first, learning, git)
├── skills/                # 13 skills (browser, testing, best practices, etc.)
├── src/                   # TypeScript source for agent authoring
│   ├── agents/            # Prompt definitions (source of truth)
│   ├── config/            # Agent metadata and routing
│   ├── adapters/          # Markdown renderer
│   └── scripts/           # Generation and validation
└── package.json           # Dependencies and scripts

Do not edit agents/*.md by hand. Update TypeScript under src/, then run pnpm generate:agents.

---

Agents

Core Pipeline

Agent	Purpose	Model
orchestrator-agent	Routes work, manages lanes and approval gates	qwen3.7-plus
research-agent	Codebase mapping, root-cause analysis	deepseek-v4-flash
planning-agent	Implementation-ready plans	gpt-5.5 (xhigh)
reviewer-agent	Adversarial plan stress-testing	glm-5.1
implementation-agent	Step-by-step plan execution	kimi-k2.6
test-fixer-agent	Diagnose and repair failing tests	deepseek-v4-pro
verifier-agent	Post-implementation audit	gpt-5.5 (xhigh)

Specialized

Agent	Purpose	Model
pr-review-agent	PR review via gh CLI	glm-5.1
prompt-agent	Refine agent prompts	qwen3.7-plus
spec-agent	Proposal + capability specs	kimi-k2.6
apply-agent	Execute playbooks (RED→GREEN)	deepseek-v4-flash
readme-architect	Professional READMEs	qwen3.7-plus
technical-product-guide	Codebase navigation guidance	qwen3.7-plus

---

Commands

Execution Lanes

Command	Use	Flow
/quick	Small, single-file changes	research → plan → implement
/medium	Multi-file work, refactors	research → plan → review → implement → verify
/full	Risky, architecture-sensitive	research → plan → review → approve → implement → verify
/debug	Bug fixing	triage → research → plan → review → approve → implement → verify

Spec-First

Command	Purpose
/spec	Create proposal.md + capability specs
/design	Generate design.md + tasks.md + implementation.md
/apply	Execute implementation playbook
/review	Multi-dimensional review of changes
/spec-flow	Full pipeline: spec → design → apply → review

Other

Command	Purpose
/learning-path	Generate structured onboarding docs
/mentor	Senior dev guidance without direct solutions
/commit	Gitmoji commit from staged changes

All lanes run through orchestrator-agent with [y/N/edit] approval gates before edits begin.

---

Skills

Skill	Purpose
agent-browser	Browser automation (navigate, forms, scrape, test)
angular-best-practices	Angular: standalone, signals, RxJS, routing
nestjs-best-practices	NestJS: modules, DI, security, performance
vercel-react-best-practices	React/Next.js performance optimization
jest-testing / karma-testing	Test authoring and stabilization
playwright-cli	Browser automation and Playwright tests
pr-review	Evidence-based PR review
caveman	Token-efficient communication mode
budget-executor / budget-explorer	Budget-aware operations
find-skills	Discover and install agent skills

---

Configuration Highlights

{
  "model": "opencode-go/qwen3.7-plus",
  "permission": { "bash": "ask", "edit": "allow" },
  "compaction": { "auto": true, "prune": true },
  "watcher": { "ignore": ["node_modules/**", "dist/**", ".git/**"] }
}

• bash: ask — requires confirmation before shell commands
• edit: allow — agents can modify files within their permissions
• Auto-compaction with pruning keeps context windows efficient
• Plugin: @warp-dot-dev/opencode-warp

---

Scripts

Script	Purpose
pnpm install	Install dependencies
pnpm check	Full check: typecheck + validate + test
pnpm generate:agents	Generate agent Markdown from TypeScript
pnpm test	Run tests with Vitest
pnpm ci	Full CI pipeline

---

Usage Examples

/quick add null check to parseUser in auth.ts
/full add dark mode toggle to settings page
/debug API returns 500 when creating orders with empty cart
/spec-flow add OAuth2 authentication with Google provider
@pr-review-agent review PR #42
/learning-path authentication module
/mentor how do I implement rate limiting in Express?

---

Agent Authoring

1. Edit files under src/agents/, src/config/, or src/adapters/
2. Run pnpm generate:agents
3. Review the agents/*.md diff
4. Run pnpm check
5. Commit both TypeScript and generated Markdown changes

Pre-commit hooks run pnpm check:agents and pnpm test automatically.