job-applier

name	job-applier
description	Create complete job application packages with tailored CV and cover letter. Use this skill whenever the user wants to apply for a job, create a job application, generate a resume/CV, or mentions "apply for this job", "job application", "tailor my resume", "cover letter for job". Automatically analyzes job requirements, researches the company, generates role-optimized resume.json, creates resume in Reactive Resume, generates PDF, and creates customized cover letter.
compatibility	Requires file system access, Python 3, and `typst` on PATH (brew install typst). Reactive Resume API key is optional and only needed if you want to also push to the server.

Job Application Skill

Build a tailored application package from a job description. See AGENTS.md for project structure.

If the user wants to find jobs first (not paste a JD), use the linkedin skill to search and extract the JD, then return here.

Workflow

1. Validate capacity

Read applications/application-tracker.json and check for duplicate company entries. Do not enforce an active-application count limit; the tracker may contain any number of active applications.

2. Parse the JD

Extract company, job title, location, work arrangement, required vs preferred skills. Infer role from the title:

• frontend — UI/UX, React, Vue, CSS
• backend — APIs, servers, databases
• fullstack — both
• data — analysis, ML/AI
• devops — cloud, infra, CI/CD

Default to fullstack when unclear.

3. Score & decide (apply/skip gate)

Don't burn effort tailoring a package you won't send. Save the JD to a temp file, then:

python3 tools/match_score.py \
  --jd "{path-to-jd}" \
  --required "Skill A" "Skill B" "5+ years X" \
  --preferred "Nice C"

Verdict rubric: 90-100% overqualified (flight-risk) · 75-89% apply · 60-74% apply with strong cover · <60% skip unless dream role.

The match score is a deterministic keyword count, not a judgment. Before trusting a low verdict, scan each MISS for adjacent tech in the candidate's profile (NestJS ≈ FastAPI, Azure ≈ AWS, Vue ≈ React) and surface those to the user — they're partial matches the score ignores. Then show the tool's output verbatim, add the adjacency notes, and ask Proceed with this application? (y/n). On skip, stop here. The red-flag scan misses context and false-positives on any $ — treat as a prompt for judgment, not a verdict.

4. Research the company

Web search for mission, recent news, tech stack, culture. Synthesize 3-5 insights you'll actually use in the cover letter. If research fails, fall back to JD-only — don't fabricate.

5. Build the package skeleton

One call creates the directory, saves the JD, scaffolds analysis.md, generates resume.json, drafts the cover letter, and upserts the tracker:

python3 tools/apply.py \
  --job "{path-to-jd}" \
  --company "{Company}" \
  --position "{Job Title}" \
  --role "{role}" \
  --keywords "keyword1" "keyword2" \
  --priority "Medium"

--keywords seeds the tailored resume (injected into the right skill group) and the analysis.md scaffold. Pass the strategic JD terms here — not every word, just the 5-10 the role hinges on.

Output lands in applications/active/{Company}/.

6. Add package to INBOX.md

After a manual apply.py package is created, add an active row to applications/INBOX.md so the package appears in the review/submission queue:

- [ ] **{Job Title}** @ {Company} (manual package) · [JD](./active/{Company}/research/job-description.md) · [CV](./active/{Company}/documents/resume.pdf) · [Letter](./active/{Company}/documents/cover-letter.md)

If you have the original application URL, append · [Apply ↗]({url}). Use URL-encoded spaces in Markdown links when needed.

7. Fill in the human parts

apply.py only scaffolds — these still need your judgment:

• research/analysis.md — match score per requirement, top 3 selling points, skill gaps with positioning. See templates/analysis-template.md for structure (already seeded into the file).
• documents/cover-letter.md — replace the scaffold with the real opening, top 3 evidence-backed matches, and tone matched to company culture. 200-400 words, 3 paragraphs.

Pull candidate background from LinkedIn-CV-Profile.md and templates/base-resume.json. Never invent experience.

8. ATS coverage check

python3 tools/ats_check.py \
  --resume "applications/active/{Company}/documents/resume.json" \
  --jd "applications/active/{Company}/research/job-description.md" \
  --critical "Skill A" "exact phrase B"

Without --critical, the tool auto-discovers the top 10 JD keywords. Exits 1 below --threshold (default 80%). Density target 2-4× per keyword.

If coverage fails, either re-run tools/resume.py --keywords <missing> to inject the term into the right skill group, or — if it's a genuine gap — leave it out and let the cover letter address it.

9. PDF render

apply.py already renders documents/resume.pdf locally via Typst — confirm it exists and inspect it. If the render was skipped (e.g. typst missing), the warning surfaces in the validation output and you can re-render with:

python3 -c "from pathlib import Path; from tools.lib.pdf import \
  render_resume_pdf_from_file; \
  render_resume_pdf_from_file(Path('applications/active/{Company}/documents/resume.json'), \
                              Path('applications/active/{Company}/documents/resume.pdf'))"

Optional: also push to Reactive Resume if the user wants the editable copy on the server — not required for submission.

python3 tools/reactive_resume.py push \
  --file "applications/active/{Company}/documents/resume.json" \
  --name "{Company} - {Job Title}" \
  --slug "{company-slug}-{YYYY-MM-DD}" \
  --tags "active" "{role}"

Add --dry-run to preview without hitting the API.

10. Validate, then hand off

apply.py already emits warnings for unresolved placeholders, off-target cover letter length, and wrong paragraph count — surface those to the user.

Then run the semantic checks in templates/quality-framework.md before handing off: truthfulness against templates/base-resume.json + LinkedIn-CV-Profile.md, tone match, no buzzword stuffing. Those are judgment calls code can't make for you.

11. Lock after submission

Only when the user confirms they've submitted:

python3 tools/reactive_resume.py lock <resume-id>

Notes

• Never fabricate experience — only what's in LinkedIn-CV-Profile.md and base-resume.json.
• Match JD terminology exactly; hiring managers and ATS both pattern-match.
• Vague JD → note missing info and make assumptions explicit.
• Duplicate slug → append a number (acme-corp-2026-03-09-2).