Resume Tailor (evidence-only, with skill confirmation)

Tailors the current CV DOCX for one job using cv_tailor.py. Claude does the comparison and asks the user which missing skills they really have; the script applies the edits and exports a PDF (the temporary DOCX is deleted). No Gemini call, so it uses none of the 20/day free quota.

Several postings at once (user gives 2+ URLs): use the batch flow below instead of repeating the steps per job.

Batch flow (several postings)

1. python cv_tailor.py jds <scratchpad>/jds <url1> <url2> ... fetches all JDs in parallel into

jd_1.txt, jd_2.txt, ... (input order). For any line marked FAILED, open the page in the browser or ask the user to paste it.

1. Read the CV(s) once (cv, and the AI CV with BASE_CV_DOCX=... if any posting is AI-focused).
2. Compare each JD (step 3 below). Ask ONE AskUserQuestion round for all jobs: merge the missing

skills across postings and ask each skill once (a confirmed skill counts for every job that needs it).

1. Write <scratchpad>/batch.json:

[{"company": "Acme Backend", "base": "main", "edits": {"replace": [...], "add_skills": [...]}}, ...] base is main (BASE_CV_DOCX) or ai (BASE_CV_DOCX_AI). Use a distinct company per posting (e.g. company + role) so the PDFs don't overwrite each other.

1. python cv_tailor.py apply-batch <scratchpad>/batch.json (optional --workers N, default 3).

All edits are checked before any PDF is made; PDFs export in parallel, one Word instance per worker.

1. Report per job as in step 7 (one table per job is fine).

Steps

1. Get the job description. Run python cv_tailor.py jd "<url>". Any job link works: LinkedIn,

Welcome to the Jungle, Ashby and Greenhouse use the site's own data for a clean description; other sites (company career pages, Workday, etc.) fall back to reading the page. If it fails (login walls, closed jobs), ask the user to paste the description.

1. Read the CV. Run python cv_tailor.py cv (prints each paragraph of the base CV DOCX).
2. Compare. List the JD's skills/requirements, in English even if the JD is French. Split them into:

• already on the CV (nothing to do, or surface better wording if the evidence is there)
• missing from the CV

1. Ask with multi-select. Use AskUserQuestion with multiSelect: true: up to 4 questions of up

to 4 missing skills each, mandatory JD skills first. Question text like "Which of these have you used in real work?". The user can type extra detail in "Other" (e.g. the tool they used). If a confirmed skill needs context to be described honestly (where/how), ask a short follow-up.

1. Draft edits (rules below) and write them to a JSON file in the scratchpad:

{"replace": [{"old": "...", "new": "..."}], "add_skills": ["..."]}. Each old must sit inside ONE run of the DOCX. Bold highlights are separate runs, so keep each edit within a plain or a bold segment of the paragraph text from step 2.

1. Apply. Run python cv_tailor.py apply "<Company>" <edits.json>.
2. Report the "Words changed" output from apply (shown as removed -> added), the confirmed skills added, the page count,

the JD skills still missing, and the output paths.

Edit rules

• Only facts on the CV or confirmed by the user in step 4. Never invent tools, numbers or domains.
• Keep content intact: never delete Projects, the NDS acquisition note, education details or the

word "fintech" in the SMARTS bullet. No "| Fintech" in the title line.

• Confirmed skills go into the Frameworks & Tools: line via add_skills, not into a specific

job's bullet, unless the user said where they used it.

• "Claude Code hooks (guardrails)" is added to the AI skills line of EVERY tailored CV by apply

(CV_ALWAYS_AI_SKILLS in config.py; AI & Upskilling: on the main CV, AI & LLM: on the AI CV). Don't put it in add_skills or a replace yourself. Mention it in the step 7 report.

• AI CV only: apply also runs CV_AI_DEFAULT_REPLACE from config.py (adds the weekly HR agent to the

job-search project bullet). Don't repeat it in your edits; mention it in the report.

• Never write "learning / deepening / upskilling in X" for a gap.
• English only. Do not switch to the JD's language.
• Going over 1 page is acceptable. Prefer edits that don't add lines, and say when the CV grows.

Output

resume_adjusted\resume_<company>.pdf only (folder from RESUME_OUTPUT_DIR in config.py). An existing file with the same name is overwritten.

Base CV: BASE_CV_DOCX in config.py (override with BASE_CV_DOCX in config.py). PDF export needs Microsoft Word.

AI-focused roles ("use AI resume"): prefix both cv and apply with the env var BASE_CV_DOCX="<path>", where <path> is BASE_CV_DOCX_AI in config.py. It has an AI & LLM: skills line; put confirmed AI skills there with a replace, not add_skills.

Legacy

resume_tailor.py (Gemini, batch from tracker) targets the OLD CV layout (fixed paragraph numbers) and a PDF base path, so it does not work with the current CV. Use the steps above.