CLI quickstart

If your product is a codebase, the fastest way to give your agent real knowledge is to let it read the code. No account setup first, no copy-pasting docs.

One command

From your project root:

npx helpshelf

That scans your project, creates your agent, and prints your embed snippet and help center URL. You'll be asked for an email address to attach the agent to.

What it reads

Category	What's extracted
UI text	Headings, button labels, placeholders, aria-labels
AI context files	CLAUDE.md, .cursorrules — product features, pricing, flows
Framework & stack	Next.js, React, Vue, Svelte, Express, Django, Rails, and others
API routes	App Router, Express, Fastify endpoints
Environment variables	Names only — never values
Dependencies	package.json, requirements.txt, Gemfile, and equivalents
README & docs	Markdown files and their structure
Functions	Signatures, parameter and return types, JSDoc
Imports	Which files depend on which
Tests	describe and it descriptions, mined as documentation
Database schemas	Prisma, Drizzle, and similar

For TypeScript and JavaScript it parses a real syntax tree rather than pattern-matching, so signatures and import graphs are accurate. If that's unavailable it falls back to regex extraction.

What it doesn't send

Environment variable values are never read or transmitted — only names, so the agent knows a STRIPE_SECRET_KEY exists without ever seeing it.

Your source code is analysed to build a structured snapshot. Review it before pushing: it's written to .helpshelf/snapshot.json in your project.

What it understood

After scanning, the CLI prints a summary — the actual headings and labels from your UI, the product features it found, your meaningful components, and the questions your agent is now equipped to answer. Read it. If the summary is wrong about your product, the agent will be too.

Then keep it current

npx helpshelf sync

Re-scans and re-pushes. Worth wiring into your release process so your agent doesn't drift from your product.

Check your score

npx helpshelf score

Reports how legible your product is to AI assistants. See AI legibility score.

Full command list

See CLI commands.