threejs-skills
Skill files that give Claude Code current, verified knowledge of three.js, so the code it writes runs on the release you actually install. The skills target three.js r186 (three@0.186).
A universal OpenCode workflow setup with specialized agents, skills, commands, and plugins for any software project.
$ npx -y skills add cloudai-x/opencode-workflow --agent claude-code
A universal OpenCode workflow setup with specialized agents, skills, commands, and plugins for any software project.
| Component | Count | What It Does |
|---|---|---|
| Agents | 7 | Orchestrator + 6 specialists (security, tests, docs...) |
| Commands | 12 | /review, /commit, /architect, /rapid, /debug... |
| Skills | 7 | Domain knowledge for APIs, testing, architecture... |
| Plugins | 5 | Auto-format, security scans, notifications... |
Your request ──► Orchestrator ──┬──► Code Reviewer ──┐
├──► Security Auditor ──┼──► Synthesized
├──► Test Architect ──┤ Report
└──► Debugger ──┘
(all parallel)
OpenCode looks for workflow files in .opencode/ inside your project directory. This repo uses different folder names (agents/ instead of agent/) so it's easier to browse on GitHub.
# Clone the workflow
git clone https://github.com/CloudAI-X/opencode-workflow.git
# Create .opencode directory in your project
mkdir -p your-project/.opencode
# Copy all components (note: folder names change!)
cp -r opencode-workflow/agents your-project/.opencode/agent
cp -r opencode-workflow/commands your-project/.opencode/command
cp -r opencode-workflow/skills your-project/.opencode/skill
cp -r opencode-workflow/plugins your-project/.opencode/plugin
Don't need everything? Pick what you want:
# Just agents and commands (no plugins or skills)
cp -r opencode-workflow/agents your-project/.opencode/agent
cp -r opencode-workflow/commands your-project/.opencode/command
| This Repository | Your Project |
|---|---|
agents/ | .opencode/agent/ |
commands/ | .opencode/command/ |
skills/ | .opencode/skill/ |
plugins/ | .opencode/plugin/ |
cd your-project
opencode
# Type / and Tab - should see commands like /review, /commit
# Press Tab repeatedly - should see Orchestrator as a primary agent
# Type @ and Tab - should see subagents like @code-reviewer
Press Tab to cycle between primary agents:
| Agent | What It Does |
|---|---|
| build | Default. Full development work. |
| plan | Analysis only, no file changes. |
| orchestrator | Coordinates complex multi-step tasks. |
Invoke specialists by mentioning them:
@security-auditor Check the auth module for vulnerabilities
@code-reviewer Review the changes I just made
| Subagent | Focus | Has Bash? | Can Edit? |
|---|---|---|---|
@code-reviewer | Quality, patterns, maintainability | No | No |
@debugger | Bug investigation, root cause | Yes | No |
@security-auditor | OWASP Top 10, vulnerabilities | No | No |
@refactorer | Clean code, design patterns | No | Yes |
@test-architect | Test strategy, coverage | No | Yes |
@docs-writer | README, API docs, guides | No | Yes |
Quick workflows for common tasks:
| Command | What It Does |
|---|---|
/review | Multi-perspective code review |
/commit | Generate conventional commit message |
/architect | High-level design session |
/rapid | Fast iteration, minimal ceremony |
/debug | Systematic bug investigation |
/refactor | Code cleanup workflow |
/security-audit | OWASP vulnerability check |
/test-design | Plan test coverage |
/docs | Generate documentation |
/parallel | Run multiple tasks at once |
/verify-changes | Lint → Type → Build → Test → Review |
/mentor | Educational explanations |
The orchestrator doesn't try to do everything itself. It follows a 6-phase workflow:
UNDERSTAND → PLAN → DELEGATE → INTEGRATE → VERIFY → DELIVER
This is the key insight: independent tasks should run simultaneously.
Five 30-second reviews take 30 seconds total, not 2.5 minutes.
The trick is putting all subagent calls in a single message. Separate messages = sequential. Single message with multiple calls = parallel.
One reviewer misses things. Multiple reviewers with different mandates catch more:
The tension between perspectives surfaces important tradeoffs.
Plugins protect without blocking your workflow:
.env, credentials, keysCreate your-project/.opencode/agent/my-agent.md:
---
description: What this agent does
mode: subagent
model: anthropic/claude-sonnet-4-20250514
tools:
write: false
edit: false
bash: false
---
Your custom instructions here...
Create your-project/.opencode/command/my-command.md:
---
description: What this command does
agent: build
---
Your command prompt here...
Use $ARGUMENTS for user input.
Create your-project/.opencode/skill/my-skill/SKILL.md:
---
name: my-skill
description: Domain knowledge for...
---
Knowledge content here...
Agents not appearing after Tab?
.opencode/agent/ exists (singular agent, not agents)--- delimitersCommands not showing with /?
.opencode/command/ (singular)commit.md → /commit)Plugins causing errors?
?.) when accessing potentially undefined propertiesSkills not loading?
SKILL.md insidename field in frontmatter should match the folder nameInspired by claude-workflow-v2 patterns. Built for OpenCode CLI.
Created by @cloudxdev
MIT
Skill files that give Claude Code current, verified knowledge of three.js, so the code it writes runs on the release you actually install. The skills target three.js r186 (three@0.186).
A universal Claude Code workflow plugin with specialized agents, skills, hooks, and mode commands for any software project. Compatible with skills.sh — works with Claude Code, Cursor, Codex, and 35+ AI agents.
opencode-workflow is a Claude Code plugin with 7 hand-picked skills, indexed on Flowy in the Agent Orchestration category. Install it with the command on its page. It includes analyzing-projects, designing-apis, designing-architecture. Its skills do not fire on their own yet. Request auto-invocation to have Flowy route them as you prompt. Free and open source.
Is this plugin yours?
Claim it with GitHubSubmit a pluginPromote it