| name | apply-to-job |
| description | Orchestrate the full apply chain for one job — capture the job once, score fit, tailor the resume (PDF), write the cover letter, and answer the application questions — in auto-pilot (run end-to-end, pause only at decision/send gates) or review (confirm each step) mode. Stops before anything that sends or submits. The single entry point for "apply to this job".
|
| when_to_use | Use when the user has a specific job (a URL, a pasted JD, or a job_id from find-jobs) and wants the complete application package assembled without running each skill by hand.
|
| user-invocable | true |
| allowed-tools | Read, Write, Bash |
apply-to-job
One command to turn a job into a ready-to-submit application package. This skill orchestrates the
other skills (it invokes them in order) — it does not re-implement them. Stage map + each skill's role:
../../knowledge/pipeline.md. Hard rules: ../_shared/RULES.md.
Input
A job, supplied any way: a URL, pasted JD text, or a job_id from jobs/found-<date>.json.
If none is given, read jobs/current.json; if that's absent too, ask for the job.
Mode
Ask once at the start (or accept --mode auto|review):
"Run this in auto-pilot (I run the whole chain and only stop at the apply/skip call and before anything is sent) or review (I check with you before each step)?"
Default to review if the user doesn't choose. Both modes obey the same guardrails below.
Chain
- Preflight (notify the user before starting). Run
python3 scripts/doctor.py. If rendercv is missing, tell the user now (it's needed for the resume PDF later in the chain) and offer the one-time install — don't discover it mid-chain. Then check profile/master-profile.md: if missing → run /build-profile first (or stop and say so).
- Capture the job once. Resolve the JD (fetch the URL / use the pasted text / look up the
job_id) and write jobs/current.json = {company, role, url, job_id, jd_text, region, source, captured_at}. Every downstream skill reads this — the JD is supplied once (RULES §6).
score-fit. Run it against jobs/current.json.
- Skip / low fit: in auto mode, stop and report the one-line reason (offer: "apply anyway?"); in review mode, ask whether to proceed.
- Apply / Apply-if-tailored: continue.
tailor-resume. Produces the tailored resume PDF (+ md + sidecars) and runs the review-render QA loop. Relay any ATS warning.write-cover-letter. Region-aware (US: role-dependent; IN: short-form often enough — follow the region pack). Skip only if the user opts out.answer-application-questions. Only if the user provides the form's questions (paste or a JSON schema). If no form is available yet, note it and move on — the answers come at submit time.- Assemble + STOP. Present the package: resume PDF path, cover letter, any answers, and the apply URL. Do not submit, send, or auto-fill any portal — that's the human's action (or the JobClaw agent's, behind its own guardrails).
In review mode, pause after each numbered step for a go/no-go. In auto mode, run 1→6 without pausing except at the score-fit decision (step 2) and the final stop (step 6); surface every artifact + warning in the end summary.
Guardrails
- Never submit, send, or auto-fill an application, email, or portal. This skill stops at a ready package.
- Never fabricate (RULES §1); never re-ask for data already captured (RULES §6 — the profile +
jobs/current.json + sidecars hold it).
- Honor the region pack and work-authorization rules; relay honesty flags from score-fit/tailor-resume rather than hiding them.
- If any step fails (e.g. rendercv not installed), continue the rest, mark that step pending, and tell the user exactly what to fix.
Output
A package summary: company · role · fit score · resume.pdf · cover-letter · answers (or "form pending") · apply URL, plus any honesty/ATS warnings.
Next steps
Application package assembled. Review it, then submit yourself (or hand it to the JobClaw agent for auto-apply). After you apply, run /triage-inbox on recruiter replies and /infer-status to track it. No emoji.
