Files

Skills for claude.ai chat

Three skills for job applications in a claude.ai chat. They hold every rule; the user's own documents hold facts only.

Skill What it does
application-drafter From one posting: finds the employer's own posting, runs a fit check (eligibility and language gates, skills, experience and career scores), builds a requirement coverage table, drafts the CV and cover letter, reviews them as a hiring manager with independent company research, applies the edits, audits, and outputs the drafts with coverage, keyword and audit tables. Also drafts application form fields. Markdown by default, .docx on request.
interview-prep For a scheduled interview on an application already sent: verified company and interviewer research, likely questions ranked by stage, STAR answers drawn only from the career master, honest bridges for gaps, a consistency brief against the submitted CV and letter, questions to ask, a mock interview with feedback, and thank-you and follow-up notes (drafts only, at most two follow-ups). Needs the posting and the submitted CV and letter besides the two documents.
application-auditor Audits any application text (CV, letter, form answer, LinkedIn post): runs scripts/check_text.py for dashes, ASCII date ranges, banned phrases, bullets per role, bullet length and character counts, then checks every claim against the career master, every pitfalls entry, the interview backtrack test and confidentiality. Returns a findings table and never fixes silently.

All three end with a "Pitfalls to add" block in the wrong / right / why format, for the user to paste into the pitfalls document. The drafter and interview-prep also list "Facts to add" for facts confirmed in the chat that the career master lacks.

The user's documents

  • Career master: facts only: roles, titles, dates, numbers, projects, education, languages.
  • Pitfalls: facts that get written wrong (a wrong title, an inflated number, a client that must stay masked), as wrong / right / why.

Paste or attach both at the start of each chat; the skills ask for them when they are missing. They also read them from a Project's knowledge if they are kept there. Nothing else belongs in either document: no rules, no style notes. Without the pitfalls document the skills still run and report the pitfalls check as skipped; the drafter does not draft without the career master.

Build and upload

./build.sh

This writes dist/application-drafter.zip, dist/application-auditor.zip and dist/interview-prep.zip (dist/ is gitignored). Each zip holds the skill folder with SKILL.md at the folder root; tests are left out. Upload each zip in claude.ai under Settings > Capabilities > Skills. Code execution must be on: the auditor runs its script in the sandbox and both skills count words and characters with code.

Run the checker's tests with python3 application-auditor/scripts/test_check_text.py.

Sources

The rules are adapted from ai-job-search (MIT, Copyright (c) 2026 Mads Lorentzen); each skill folder carries the licence in NOTICE. Its LaTeX templates, application tracker, salary tool and Claude Code mechanics (subagents, file writes, caches) are not ported.

Planned

An upskilling skill (a gap analysis and study plan for one posting) may follow.