From onboard
Onboard to any codebase - generates a structured overview then guides you interactively through the project
npx claudepluginhub agent-sh/agentsys --plugin onboard[path] [--depth=quick|normal|deep]# /onboard - Codebase Onboarding Onboard to any codebase. Collects project data automatically (no LLM), then an agent synthesizes it and guides you interactively. ## Arguments Parse from `$ARGUMENTS`: - **Path**: Directory to analyze (default: current directory) - `--depth`: Analysis depth - `quick`: Manifest + README + directory tree only (~2s) - `normal` (default): + CLAUDE.md, CI, repo-intel (~5s) - `deep`: + repo-map AST symbols (~15s) ## Phase 1: Automated Data Collection (Pure JS) ## Phase 2: Agent Synthesis + Interactive Guidance
/onboardGenerates four audience-tailored onboarding guides—Contributor, Staff Engineer, Executive, Product Manager—in an onboarding/ folder with index hub.
/onboardGenerates a developer onboarding guide by scanning project configs, structure, and docs for prerequisites, setup, workflow, key concepts. Writes to docs/onboarding.md.
/onboardScans new codebase, checks context and files, classifies stack, maps architecture, sets up rules, and saves knowledge to .memory/ for building.
/onboardInitiates guided KYC customer onboarding workflow: enforces stagegates, gathers details, creates case folders with consent, runs verifications, generates Excel/PDF reports.
/onboardGenerates customized onboarding plans for new hires based on role, start date, location, tech requirements, and team details. Covers pre-arrival prep, Day 1 setup, and first 90 days.
/onboardGenerates codebase onboarding reports: architecture walkthroughs, key files, naming analysis, dependency maps, and guided code tours. Outputs Markdown files to docs/onboarding/ and creates a commit. Supports flags like --quick, --tour.
Onboard to any codebase. Collects project data automatically (no LLM), then an agent synthesizes it and guides you interactively.
Parse from $ARGUMENTS:
--depth: Analysis depth
quick: Manifest + README + directory tree only (~2s)normal (default): + CLAUDE.md, CI, repo-intel (~5s)deep: + repo-map AST symbols (~15s)const { getPluginRoot } = require('@agentsys/lib/cross-platform');
const pluginRoot = getPluginRoot('onboard');
const collector = require(`${pluginRoot}/lib/collector`);
const args = '$ARGUMENTS'.split(' ').filter(Boolean);
const depth = args.find(a => a.startsWith('--depth='))?.split('=')[1] || 'normal';
const targetPath = args.find(a => !a.startsWith('--')) || process.cwd();
console.log(`[INFO] Collecting project data (depth: ${depth})...`);
const data = collector.collect(targetPath, { depth });
console.log(`[OK] Data collected:`);
console.log(` Manifest: ${data.manifest?.type || 'none'} (${data.manifest?.language || '?'})`);
console.log(` Structure: ${data.structure?.length || 0} directories`);
console.log(` README: ${data.readme ? 'found' : 'missing'}`);
console.log(` Repo-intel: ${data.repoIntel ? 'available' : 'unavailable'}`);
console.log(` Repo-map: ${data.repoMap ? data.repoMap.totalFiles + ' files, ' + data.repoMap.totalSymbols + ' symbols' : 'unavailable'}`);
await Task({
subagent_type: "onboard:onboard-agent",
prompt: `Onboard the user to this codebase.
## Collected Data (already gathered, do NOT re-scan files)
${JSON.stringify(data, null, 2)}
## Your Job
1. **Synthesize** the collected data into a clear, concise summary (2-3 min read)
2. **Read key files** that the data points to - entry points, main modules, interesting patterns
3. **Present the summary** to the user
4. **Ask what they want to do** - fix a bug? add a feature? understand a specific area?
5. **Guide them** to the right files using coupling, ownership, and symbol data
Use the repo-intel onboard data to enrich the summary:
- Health and bus factor
- Key areas with purpose annotations
- Pain points (high bug-fix rate areas)
- Hotspots (where active development is)
- Getting started commands
Use conventions data (if available) to describe coding style:
- Function naming convention (snake_case, camelCase, PascalCase)
- Test framework and test location pattern
- Commit message convention (conventional, freeform)
Use project metadata (if available) for quick facts:
- Primary language and language breakdown
- CI provider and workflow count
- License type
- Package manager
Use repo-map data (if available) to trace code:
- "This function is exported from X and imported by Y and Z"
- "The main entry point calls these modules in this order"
Do NOT just dump the JSON. Synthesize it into human-readable insights.
After the summary, ask the user what they want to explore.`
});