Skip to content

🇺🇸 English · 🇧🇷 Português

Command line

Every slice also installs a command you can drive from a shell or a cron job, with no LLM conversation involved. The CLIs don't depend on Claude as an MCP client; the steps that score or draft still call the LLM backend in config.yaml, and --no-eval makes a scan with no LLM call at all.

Output and exit codes

Every moonlighter-* command, and moonlighter doctor, prints exactly one JSON document on stdout (logs go to stderr) and exits — bare moonlighter is the MCP server over stdio and moonlighter init is an interactive wizard, so neither prints JSON:

Exit Meaning
0 success
1 nothing to do (no new jobs, job not found, no questions)
2 usage or config error
3 unexpected error — the JSON then carries kind: "error" and the traceback goes to stderr

Commands

Command What it does
moonlighter-scan [--phase all] [--keywords ...] Run a scan; --company SOURCE SLUG scans one board. --no-eval discovers and stores postings as needs_review without calling the LLM — score them later with verify_job.
moonlighter-apply prepare JOB_ID [--paste FILE] Compose the paste-ready sheet; --paste - reads the page text from stdin.
moonlighter-apply prepare --url URL [--company X --title Y] [--paste FILE] Ingest the posting first: through its ATS API when the URL has a known shape, otherwise from the page itself with --company and --title supplied; stored unscored, then prepared. No LLM call for the ingest.
moonlighter-apply bootstrap-cv [--force] [--skip] Draft a CV pool and template from profile.yaml — see Tailored CV.
moonlighter-email sync Classify recent replies and advance applications. Standalone it does not feed the answer bank; the MCP server's sync_email_responses does.
moonlighter-email register JOB_ID Mark a job as applied by hand and mint its +ref tracking alias, so replies to it are matched by sync.
moonlighter-scan doctor · moonlighter-apply doctor · moonlighter-email doctor · moonlighter doctor Where the state lives and whether the config loads, as JSON; exit 1 when the config is missing or invalid. 
moonlighter-scan --no-eval | jq '.saved[] | select(.status == "needs_review") | .url'

The example above exits 1 on every quiet day (no new jobs), which trips set -e/pipefail in a script that chains it with jq — check the exit code before treating that as a script failure. --no-eval is zero-LLM, not offline: archive_stale_jobs still makes HTTP requests to check whether previously-saved jobs closed.

What each install gives you

The five packages are slices of one tool. Install the ones you need; each command tells you in --help what it can do here and what a missing slice would add, and doctor prints the same as JSON.

Each moonlighter package works alone; installing two lets them work together

Each package works alone, and installing two lets them work together: scan results feed prepare, and the +alias a sheet mints is the one sync matches replies against. moonlighter installs the three plus moonlighter-core. Every moonlighter-* command prints one JSON document, whatever you install.

You install You get
moonlighter-scan moonlighter-scan: boards and portals scanned, postings scored (or stored unscored with --no-eval), closed ones archived
moonlighter-apply moonlighter-apply prepare: the paste-ready sheet, from a job id or straight from a URL
moonlighter-email moonlighter-email register and sync: applications registered by hand, Gmail replies matched back to them
moonlighter-scan + moonlighter-apply one script: scan, pick by score, prepare a sheet for each
moonlighter-apply + moonlighter-email the tracking alias a sheet mints is the one sync matches replies against
moonlighter (everything) all of the above plus the MCP server for Claude Code, and answer-bank promotion when a reply advances an application 
moonlighter-apply doctor | jq '.slices, .capabilities.missing[].name'

← Back to the README