Multi-agent orchestration harness for OpenCode. One install, complete control.
# One-time setup
ocx init --global
# Add the registry
ocx registry add https://fahrigedik.github.io/workspace --name fahrigedik
# Install the workspace
ocx add fahrigedik/workspace --global
# Launch
ocx oc -p workspaceOr install directly without saving the registry:
ocx add fahrigedik/workspace --from https://fahrigedik.github.io/workspace --globalA bundle — a curated collection of components that work together as a complete AI development harness. Installing fahrigedik/workspace gives you:
- 4 plugins (delegation, planning, notifications, worktrees)
- 5 agents (researcher, coder, scribe, reviewer, architect)
- 6 skills (plan protocol, code review, code philosophy, frontend philosophy, testing philosophy, plan review)
- 1 command (/review)
- Orchestrator configurations for plan/build/explore agents
- Permission boundaries (webfetch deny, agent sandboxing)
┌──────────────────────────────────────────────────────────┐
│ ORCHESTRATORS │
│ ┌──────┐ ┌────────┐ ┌───────┐ │
│ │ plan │ │ review │ │ build │ │
│ └──┬───┘ └───┬────┘ └───┬───┘ │
└────────┼────────────┼─────────────┼────────────────┘
│ │ │
┌─────┴─────┐ ┌──┴──┐ ┌──────┴──────┐
▼ ▼ ▼ ▼ ▼ ▼ ▼ ▼
┌─────────────────────────────────────────────────────────┐
│ SPECIALISTS │
│ ┌──────────┐ ┌──────────┐ ┌───────┐ ┌──────┐ ┌──────┐ │
│ │ explorer │ │researcher│ │ coder │ │scribe│ │review│ │
│ └──────────┘ └──────────┘ └───────┘ └──────┘ └──────┘ │
│ ┌──────────┐ │
│ │ architect│ │
│ └──────────┘ │
└─────────────────────────────────────────────────────────┘
| Role | Agents |
|---|---|
| Orchestrators | plan, review, build |
| Specialists | explore, researcher, coder, scribe, reviewer, architect |
| Category | Component | Description |
|---|---|---|
| Plugin | workspace-plugin | Plan management, agent rule injection |
| Plugin | background-agents | Async delegation system |
| Plugin | notify | OS notifications on completion |
| Plugin | worktree | Git worktree isolation |
| Skill | plan-protocol | Implementation planning guidelines |
| Skill | code-review | Review methodology + severity classification |
| Skill | code-philosophy | Internal logic philosophy (5 Laws) |
| Skill | frontend-philosophy | Visual/UI philosophy (5 Pillars) |
| Skill | testing-philosophy | Testing standards and methodology |
| Skill | plan-review | Plan quality assessment |
| Agent | researcher | External research (MCP tools, read-only) |
| Agent | coder | Implementation (full file + bash) |
| Agent | scribe | Documentation (write, no bash) |
| Agent | reviewer | Code review (read-only + git) |
| Agent | architect | System design (read-only, proposals) |
| Command | review | /review slash command |
| Scope | Setting |
|---|---|
| Global | webfetch: deny — no direct web fetching |
| plan | Read-only orchestrator, delegates via task tool |
| review | Read-only orchestrator, PR review via gh CLI |
| build | Read-only orchestrator, delegates via task tool |
| explore | Read-only specialist, filesystem + git inspection only |
| researcher | Read-only, MCP tools only |
| coder | Full file + bash access |
| scribe | File write only, no bash |
| reviewer | Read-only + git inspection |
| architect | Read-only, analysis and design proposals only |
src/
├── plugin/ # TypeScript plugins
│ ├── workspace-plugin.ts # Plan management & rule injection
│ ├── background-agents.ts # Async delegation system
│ ├── notify.ts # OS notifications
│ ├── worktree.ts # Git worktree isolation
│ ├── workspace-utils/ # Shared utilities
│ ├── notify/ # Notification backends
│ └── worktree/ # Worktree sub-modules
├── agent/ # Agent definitions (Markdown)
│ ├── coder.md # Implementation specialist
│ ├── researcher.md # External research
│ ├── reviewer.md # Code review expert
│ ├── scribe.md # Documentation writer
│ └── architect.md # System design specialist
├── skills/ # Skill definitions
│ ├── code-philosophy/ # 5 Laws of Elegant Defense
│ ├── frontend-philosophy/ # 5 Pillars of Intentional UI
│ ├── code-review/ # Review methodology
│ ├── plan-protocol/ # Planning format
│ ├── plan-review/ # Plan quality assessment
│ └── testing-philosophy/ # Testing standards
├── command/ # Slash commands
│ └── review.md # /review command
└── philosophy/ # Philosophy documentation
└── AGENTS.md # Mandatory philosophy rules
Agent definitions are in src/agent/*.md. Each file has:
- YAML frontmatter —
descriptionandmode: subagent - Role description — What the agent does
- Authority — What it can/cannot do autonomously
- Process — Step-by-step workflow
- Forbidden actions — Hard boundaries
Edit the Markdown directly. Changes take effect on next session.
Skills are in src/skills/*/SKILL.md. Agents load them on demand via skill tool.
code-philosophy— Edit to match your team's coding standardsfrontend-philosophy— Edit to match your design systemcode-review— Add/remove review layers or severity criteriatesting-philosophy— Customize testing standards
- Create
src/agent/your-agent.md - Add YAML frontmatter:
descriptionandmode: subagent - Define clear authority boundaries and forbidden actions
- Update your OCX profile to include the agent
- Create
src/skills/your-skill/SKILL.md - Add YAML frontmatter:
nameanddescription - Reference the skill in relevant agent definitions
| Problem | Solution |
|---|---|
| Agent not found | Verify agent .md file exists in src/agent/ and has correct frontmatter |
| Delegation failed | Check that the target agent's mode: subagent is set correctly |
| Notifications not appearing | Check config at ~/.config/opencode/notify.json, verify quiet hours |
| Worktree creation fails | Ensure clean git state, check branch name validity |
| Skill not loading | Verify SKILL.md exists in src/skills/{name}/ with correct frontmatter |
OpenCode icin cok ajanli orkestrasyon sistemi. Tek kurulumla 5 ajan, 6 beceri, 4 plugin ve 1 komut icerir.
ocx init --global
ocx add fahrigedik/workspace --global
ocx oc -p workspaceSistem iki katmanlidir:
- Orkestrator ajanlar (plan, build) — isi planlar ve dagitir
- Uzman ajanlar (coder, researcher, reviewer, scribe, architect) — isleri yapar
Her ajan kesin yetki sinirlarina sahiptir. Coder kod yazar ama commit edemez. Reviewer inceler ama dosya degistiremez. Researcher arastirma yapar ama dosya sistemine dokunamaz.
Bu paket sizin. Agent'lari, skill'leri ve plugin'leri dilediginiz gibi duzenlayebilirsiniz:
src/agent/*.md— Ajan tanimlarisrc/skills/*/SKILL.md— Beceri tanimlarisrc/plugin/*.ts— Plugin'ler (TypeScript bilgisi gerekir)
Based on the KDCO Workspace by Kenny. Extended with additional agents, skills, and improvements.
MIT — Fahri Gedik