Run the llmscout CLI (published as llmscout-cli on npm and PyPI) and return its structured output as JSON. Call this to scaffold an SEO/GEO config for a local project, run llmscout's 21-check technical-SEO and generative-engine-optimization audit against a site, or batch-run that audit across a fleet of sites, without shelling out yourself.

Call it once you have a local project directory path, and for check/fleet a site that already has an llmscout.json (created by a prior init call, or already present in the project). Do not call check on a directory that has never been init-ed: it fails fast with a missing-config error rather than guessing a site URL. No API key or auth is required. check and fleet make live outbound HTTP requests to the target site(s) (the page itself, robots.txt, sitemap.xml, and related resources) so they need network access and will be slower or noisier against unreachable hosts; init only writes local files (a config and a small skill file) and makes no network calls. All subcommands are safe to re-run: init overwrites its scaffolded config, and check/fleet never write anything unless --out-dir is given, in which case a report file is rewritten each run. On failure (non-zero exit, bad args, unreachable site), the underlying process's stderr is captured rather than raised.

args is a list[str] of the CLI's own argv, split exactly as you would type it on a command line (never a single shell string), and should never include --json yourself since this wrapper appends it automatically. Real examples:

• ["init", "./my-site", "--site-url", "https://example.com"]

• ["check", "./my-site", "--out-dir", "./reports"]

• ["fleet", "./fleet.json", "--out-dir", "./reports"] Pass ["--help"] or ["", "--help"] (e.g. ["check", "--help"]) as args to discover the live, authoritative list of subcommands and flags.

Returns a dict. On success it is the parsed JSON the CLI printed: init returns the scaffolded config paths, check returns {siteUrl, summary: {pass, warn, fail, total}, results: [{id, name, category, status, message, fix?}, ...]}, and fleet returns one such result per site. On failure it returns {"error": ..., "stderr": ..., "command": ...} if the CLI exited non-zero, or {"error": ..., "stdout": ..., "command": ...} if stdout was not valid JSON; it never raises.