evaluate-job

Use when the user asks "should I bother applying", "is this worth it" or "should I still apply" about a saved job, or new information comes in. Re-checks fit, start timing, each requirement, gaps and

SKILL.md

GitHub

/evaluate-job

Usage: /evaluate-job <job link, pasted posting, or job folder name> (or just ask in plain English).

Input: $ARGUMENTS (a URL, pasted text, or an existing jobs/ folder). If the job isn't in the tracker yet, use /add-job instead: it saves the posting with all its details (deadline, job ID, apply link, required documents), runs this same evaluation, and preps the application. Use this skill to re-evaluate a job that's already saved.

  1. Get the posting. Use the job folder's posting.md. If the posting changed or the user gives a newer version, fetch the URL or use the pasted text (login wall → ask the user to paste it), update posting.md keeping its header (format in /add-job step 2), and note what changed. If there's no folder yet, this is a new job: run /add-job instead.
  2. Read profile/master-resume.md, profile/preferences.md, profile/story-bank.md.
  3. Write jobs/<folder>/fit.md:
    • Verdict line: Fit 4/5: Apply (priority A), or Skip with the reason.
    • Requirements table: every required and preferred qualification → Required / Preferred (as the posting says) → Met / Partly / Not met → the exact evidence from the master resume (quote it) or the gap. Mark knockouts (a license or certification the job legally needs, clearance, work authorization, a working language, a location the user can't meet): a missing knockout caps the score at 2. How to sort requirements and judge years-of-experience gaps: references/reading-postings.md (sections 1-2).
    • Keywords: the posting's important terms; which ones the user truthfully has (check knowledge/field/resume-keywords.md for the evidence behind each; if it still says STATUS: not generated, check the master resume directly) vs. not.
    • Gaps and how to handle them: e.g., "No Python: mention the Excel/VBA automation; don't claim Python."
    • Red flags: level mismatch against the user's career_stage (e.g., "3+ years required" for a new grad, or a junior title for a senior candidate: often still worth applying if no knockout is missing and most required items are met with evidence; say so), vague duties, reposted for months, staffing agency, scam signals from knowledge/job-search-playbook.md. Phrases like "wear many hats" or "fast-paced" become questions for the user to ask (reference, section 4), not score cuts. Take-home that looks like real work (the company's actual backlog, a live client or customer problem, real data), runs past about a day unpaid, has a vague spec, or comes with no interview around it: flag it and see .claude/skills/interview-prep/references/take-home-assignments.md §3.
    • Real? One line: Real? likely / unclear / doubtful: <why>, from posting age and "reposted" signals, whether the same req is still live on the employer's own careers site or job board (only on aggregators = yellow flag), a missing pay range where the location's pay-transparency law requires one, vague duties, and scam signals. A signal, not a verdict. Judge the combination (reference, section 3): big employers keep "evergreen" postings up all year, so age or one repost alone isn't enough for doubtful. For unclear, suggest asking a contact or named recruiter whether it's actively hiring.
    • Way in and timing: anyone at the company in contacts/contacts.csv, or a former employer? Note it (referred applicants are more likely to be hired), plus any alumni count, dated (.claude/skills/network/references/alumni.md; an alum is a way to ask, not a referral); if priority A and nobody, suggest /network <company>. If the posting says rolling review, "may close early", or caps applications, say it's urgent.
    • Work authorization: if profile/preferences.md says the user needs sponsorship, quote the posting's sponsorship language ("unable to sponsor", "must be authorized without sponsorship", US citizenship or clearance required) and flag a mismatch. For US employers, past H-1B sponsorship can be checked on the USCIS H-1B Employer Data Hub (https://www.uscis.gov/tools/reports-and-studies/h-1b-employer-data-hub); cite it as a signal only. For UK jobs that need sponsorship, follow references/visa-checks.md (the Register of licensed sponsors, and whether the job and pay look like they clear the Skilled Worker thresholds, checked against gov.uk on the day). If a visa clock is set in preferences (visa_clock_start), work out the days left with the formula in references/visa-checks.md ("Visa clock") and compare them with the employer's likely timeline and say so plainly ("about 40 days left; this process usually takes 4-8 weeks"). These are signals for the user's own decision, never immigration advice: for their situation, point them to their school's DSO (students) or an immigration lawyer.
    • Pay vs. floor: compare the posting's pay_range with the salary floor in profile/preferences.md for the job's country (the floor line may list one figure per country or currency, e.g. "US: $70,000; Germany: EUR 55,000"). Compare like with like: same currency, both annual (hourly × about 2,080 in the US; monthly × 12, or × 13-14 where a 13th/14th month salary is stated), both gross base. If the floor is only in another currency, convert at a current rate from a named source (e.g. the central bank's reference rate) with its date, say it's approximate, and mention cost-of-living differences rather than treating the converted figure as the floor. If no range is posted (or the floor is "use market range"), get a sourced market range per .claude/skills/offer/references/pay-data.md (reuse a recent "Market pay" block in this folder) and save the block in fit.md. Flag it if the top of the range sits below the floor; never guess a number.
    • Hourly, contract or academic roles (read the matching file in the playbook's "Special situations"): hourly → compare the shifts and hours with the user's availability and deal-breakers, required certifications as knockouts, the hourly rate against an hourly floor, and the hours needed for benefits; contract → length, rate on the same basis as the floor, W-2/1099 or IR35 status, contract-to-hire odds as a question for the recruiter; academic → fit with the department's research and teaching needs, every document the posting requires (statements, letters, writing sample) and how long they'll take before the deadline.
    • Company snapshot: 3 bullets (what they make or do, size, anything recent), each with a source link or marked "not verified"; use the company-researcher agent if needed. For priority A and B, if a company-data app is connected, add its funding, headcount trend, layoffs and deals as fit and risk signals with Source: <app>, read <date> (knowledge/data-connectors.md; you look it up and pass it to the agent, which has no connector tools).
    • Best stories for this job: 2–3 from the story bank.
    • Angle: one sentence on how the user should pitch themselves for this role.
  4. Start-date check (do it before you score): the user can start on the earliest start date in profile/preferences.md (for students and new grads, usually tied to graduation). If that date is now or soon, most postings pass; skip the rest of this check. If it's months away: a posting that names a matching start (new grad, university/campus recruiting, development or rotational program, "start summer <year>") passes. A posting that clearly needs someone before the user can start ("immediate start", "start within 30 days", or the recruiter said so) can't wait for them: cap the score at 2 and say so plainly. If the timing just isn't stated, don't assume: score it on fit, mark the timing "unknown, ask the recruiter", and remember the closer it gets to the user's start date, the more ordinary postings become realistic. For a clearly hire-now role at a former employer (or, for students, a former co-op/internship employer) or where they have a referral, skip the cap and suggest asking that contact about openings that fit their start date instead of applying. Scoring guide: 5 = meets nearly all required + matches their preferences; 4 = most required, minor gaps; 3 = stretch but plausible; 2 = major gaps, a missing knockout, poor preference fit, or can't wait for the user's start date; 1 = wrong level/field or red flags. Preferred items move the score within a band; they never sink it on their own. Call: 4-5 apply. 3: apply if the gap can be explained in one honest sentence and Real? isn't doubtful, or the user has a way in; otherwise skip. If the newest interviews/weekly-*.md has a Fit check: line (how fit scores matched real replies), let it tip a borderline call and say so ("your fit 3 jobs have been replying, so this one's worth a try"). 1-2 skip. The user can always override. Priority (used by /add-job too): fit 5, or fit 4 with a way in (a contact, referral or former employer) = A; other fit 4 = B; fit 3 = C. The user can change it.
  5. Tracker: re-read jobs/tracker.csv right before the edit and change only this row. If the job is already applied or further along, don't touch its status or next step; just add the fit and any new information to notes. Otherwise add or update the row: fit, priority, next_action = Review package and submit (if applying) or Decide: apply or skip, and always a next_action_date (3 days from today for A/B, 7 days for C, the same windows as /add-job; for a fit 4-5 job that closes early or reviews on a rolling basis, tomorrow at the latest; keep an existing earlier date; and never later than 3 days before the deadline; if that's already passed, today; if the deadline itself has passed, say so and suggest skipped unless the posting is still open). If the user decides to skip it, set status=skipped, clear the next step, and add a short note. Then add one line to history/activity.md.
  6. Tell the user the verdict in 3 lines and offer the next step (/apply <folder>).