OpenKit

A plugin kit for OpenCode with 24 commands, 26 skills, and 7 subagents for AI-assisted development.

Commands | Skills | Agents | Installation

---

Drop this kit into any project and get a full suite of AI development workflows — from feature planning to implementation, verification, and archiving.

Installation

Install the kit:

npm install -g @ndhuy05/openkit

Initialize OpenKit in your existing project:

openkit init

Note

Restart OpenCode after initializing to see the new commands in the slash menu.

Prerequisites

• OpenCode installed and configured
• Node.js >= 18

Optional tools

Some skills use these tools when available:

• OpenSpec: npm i -g @fission-ai/openspec@latest — spec-driven workflow (proposal, tasks, verify, archive)
• GitNexus: npm i -g gitnexus — blast radius analysis and dependency tracing

Commands

All commands use the ok! prefix. Type /ok! in OpenCode to see the full list.

Planning

Command	What it does
/ok!feat	Plan a new feature — explore, brainstorm, then implement
/ok!fix	Investigate and fix a bug — trace root cause, plan fix
/ok!chore	Plan maintenance work
/ok!refactor	Plan code refactoring
/ok!optimize	Plan performance optimization

Implementation

Command	What it does
/ok!apply	Implement tasks from a spec or conversation plan
/ok!yolo	Autonomous pipeline — explores, implements, verifies without stopping
/ok!proposal	Create an OpenSpec change with proposal, design, and tasks

Verification

Command	What it does
/ok!verify	Verify implementation matches spec — report only, no fixes
/ok!archive	Archive a completed change

Research & Analysis

Command	What it does
/ok!research	Search the web for technical info, best practices, docs
/ok!analyze	Trace dependencies, blast radius, call chains via GitNexus
/ok!explain	Explain how a code area works using Feynman Technique
/ok!database-design	Database schema design, ORM selection, indexing strategy

Tooling

Command	What it does
/ok!git	Git operations — commit, push, merge, rebase, changelog
/ok!test	Plan and implement tests
/ok!docs	Plan and implement documentation
/ok!ci	Plan CI/CD pipeline changes
/ok!docker	Plan Docker/containerization work
/ok!browser	Reproduce bugs or run QA flows in the browser
/ok!setup	Set up a project from boilerplate or tech stack
/ok!port	Port a feature from another codebase to this project
/ok!uiux-design	UI/UX design analysis and recommendations
/ok!create-readme	Generate a README for the project

Skills

Skills define the behavior behind each command. They live in .opencode/skills/*/SKILL.md.

Planning commands (feat, fix, chore, refactor, optimize) all load the shared explore skill, which provides:

• Orchestrator identity gate — planning commands never write code directly
• Feynman-first approach — restate requirements in simple language to find gaps
• Stress-test protocol — self-check questions resolved by exploring the codebase
• Zero-fog checklist — no "probably" or "should work" allowed before implementation
• OpenSpec awareness — integrates with spec-driven workflows when available

Agents

Subagents handle specialized tasks. They run in foreground, have no conversation history, and receive all context from the invoking skill.

Agent	Role
osf-apply	Writes code, completes tasks, auto-verifies
osf-proposal	Creates OpenSpec artifacts (proposal, design, tasks)
osf-verify	Independent verification — reports issues, never fixes
osf-archive	Finalizes completed changes
osf-analyze	Structural analysis via GitNexus knowledge graph
osf-researcher	Web research with cited sources
osf-uiux-designer	UI/UX design analysis and reports

Project Structure

.opencode/
  agents/          7 subagent definitions
  commands/        24 slash commands (ok!*)
  skills/          26 skill folders with SKILL.md + resources
  package.json     depends on @opencode-ai/plugin

Workflow

A typical workflow looks like:

/ok!feat "add user auth"     Plan the feature (explore mode)
/ok!apply                    Implement from the plan
/ok!verify                   Verify the implementation
/ok!archive                  Archive the completed change

Or skip straight to autonomous mode:

/ok!yolo "add user auth"     Explore + implement + verify, no stops