docs · getting started
Quickstart
From zero to a skill-backed run in about two minutes. You need Node (with npx available) and an agent that can drive a browser — Playwright MCP, claude-in-chrome, or similar.
1 · Install the CLI
npm i -g bmem-cli
2 · Bootstrap your agent
Run this once per agent. It registers a meta-skill that teaches your agent the loop: search the catalog before exploring a site by hand, how to read a SKILL.md, and how to map each skill capability (navigate, evaluate, …) to its own browser tools:
bmem install
3 · Find a skill
bmem search linkedin people # by keywords bmem search --site linkedin.com # by site bmem list --limit 50 # browse everything
Results are tiny index entries — name, intent, side effect, the capabilities it needs, and whether it is verified. Want the full recipe before committing? bmem show linkedin.com/search-people prints the SKILL.md without installing anything.
4 · Add it
bmem add linkedin.com/search-people
The skill now appears in your agent’s own skill list — you don’t call bmem again to run it. Registration delegates to the standalone skills installer, which detects your agent (Claude Code, Cursor, …) and writes to the right place.
Only add a name that came back verbatim from search or list. The catalog is curated and read-only — a made-up id will just 404.
5 · Run it
Ask your agent for the task — it invokes the skill natively and executes it with its own browser:
If the skill needs a logged-in session (auth: session) and the site isn’t logged in yet, navigate there, log in manually once, and re-run — the session persists in your browser profile. The full auth rules are on the skills page.
Next steps
- Architecture — how a skill runs and why replays stay stable, with diagrams.
- bmem CLI reference — all commands, flags and environment variables.
- Skills & SKILL.md — the file format and the capability mapping for your browser.