# @skillit/client > The `skillit` command-line interface: generate, audit, and refine AI-agent > skills from a TypeScript codebase, plus the Anthropic model client that powers > the headless refine loop. `@skillit/client` ships the `skillit` binary. It ties together the skillit source extractors (`@skillit/cli`, `@skillit/mcp`, `@skillit/core`) behind one CLI, and provides the Anthropic-backed `ModelClient` used by `skillit refine` when run without an agent. ## Commands | Command | What it does | | ---------------- | ---------------------------------------------------------------------------------------------------- | | `skillit gen` | Deterministically (re)generate the skill from current source — no model, no install. | | `skillit audit` | Audit + judge the generated skill; `--json` emits the full report with per-finding source locations. | | `skillit init` | Detect the project and install the matching `@skillit/*` package (install/wire only). | | `skillit refine` | Headless audit → draft → review loop that writes routing annotations back into source. | | `skillit mcp` | Extract / bundle skills from a live MCP server. | `skillit init` now also scaffolds a `skillit.config.ts` file (if none exists), so you can set a global `skillDir` and per-source (`cli` / `config` / `mcp` / `typedoc`) overrides. ## Quick start ```bash # Install npm install --save-dev @skillit/client # Generate a skill from a Commander CLI skillit gen --source cli --program ./dist/cli.js#program # Check its grade and findings as JSON skillit audit --source cli --program ./dist/cli.js#program --json ``` For the agent-driven workflow (recommended), use the bundled `/skillit-bootstrap` skill. This package includes the skill files, and generated CLI packages wired by `skillit init` install bundled skills to user roots via postinstall: Claude (`~/.claude/skills`), Copilot (`~/.copilot/skills`), and Codex (`~/.agents/skills`). ## License MIT