onboard-plugin
v0.1.1
Published
Interactive project onboarding for AI coding agents. Generates CLAUDE.md, SPEC.md, TASKS.md through structured interviews.
Maintainers
Readme
Onboard-plugin
Interactive project onboarding for AI coding agents. Generates agent-ready documentation (CLAUDE.md, SPEC.md, TASKS.md) through structured developer interviews.
Quick start
# Install into your project (no global install needed)
cd your-project
npx onboard-plugin init
# Then start your agent and run the onboarding
claude # or: opencode
/onboardWhat it does
When you run /onboard in Claude Code or OpenCode, the agent:
- Scans your project for existing files and tech stack signals
- Interviews you through 6 phases (vision, features, tech, architecture, rules, tasks)
- Generates four markdown files:
PROJECT_BRIEF.md— Why this exists and for whomSPEC.md— Features, tech stack, architecture decisionsCLAUDE.md— Agent instructions, coding conventions, build commandsTASKS.md— Phased development plan with checkboxes
- Validates consistency across all docs
Re-run /onboard anytime to refine — it detects existing docs and only asks
about what's changed.
Installation options
# Install for both Claude Code and OpenCode (default)
npx onboard-plugin init
# Claude Code only
npx onboard-plugin init --claude-only
# OpenCode only
npx onboard-plugin init --opencode-only
# Overwrite existing files
npx onboard-plugin init --forceStandalone commands
# Scan project state (useful for debugging)
npx onboard-plugin detect
# Validate generated docs
npx onboard-plugin validate
# Remove plugin files (keeps generated docs)
npx onboard-plugin uninstallHow it works
The plugin installs a skill directory (.skills/onboard/) and a slash command
into your agent's command directory. The skill contains:
SKILL.md— Overview the agent reads on every invocationreference.md— Full interview protocol (loaded on demand)examples.md— Sample outputs for quality calibration (loaded on demand)scripts/detect.sh— Project scannerscripts/validate.sh— Doc consistency checker
The agent reads the skill, runs detection, interviews you, and writes docs. No API keys, no cloud services, fully offline.
Extending
Add new project types by editing .skills/onboard/reference.md — add a
"Defaults by project type" section. The interview structure stays the same;
only the suggestions change.
License
MIT
