agentleFS
HimadriTryingHimadriTrying/ai-job-search-agent/.claude/skills/cv-tailor/SKILL.md

cv-tailor

Skill1 starsChanged 3 months agoMITView on GitHub

Rewrite the master CV for one specific job description. Selects and reorders real content, tailors to the posting, runs the fresh-context Reviewer, then the honesty gate. Never invents. On demand only.

13 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 13 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not fetch the page. Resolve the link to its board's public JSON instead:

    line 21
  2. Distinctive:

    Relevance-weighted cutting: when it overflows 2 pages: score each line by (a) relevance to this posting, (b) uniqueness in the doc, (c) …

    line 50
  3. Very specific:

    Every claim traces to career_facts.yaml.

    line 75

“Common” is measured: kept by at least 6 in 10 of the 13 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

cv-tailor — targeted CV (drafter half)

Precedent

A drafter writes; a fresh-context reviewer critiques; the drafter revises. Output is compiled and visually checked before it is shown.

Inputs

profile/08-cv-notes.md (the house spec: read it in full before drafting), profile/05-cv-source.md, profile/03-writing-style.md, career_facts.yaml, the target JD as a file (the gate's --job flag needs it).

Getting the JD from a URL — try this before asking for a paste. Most job pages are an ATS board rendered client-side, so fetching the HTML returns a shell with no posting in it. Do not fetch the page. Resolve the link to its board's public JSON instead:

python .claude/skills/job-scout/scout/joburl.py <url> data/jd/<company>-<role>.md

Covers Greenhouse, Lever, Ashby and SmartRecruiters, hosted (jobs.ashbyhq.com/acme/...) or embedded in a company's own careers page (acme.com/careers?ashby_jid=...). No key, no login. Exits non-zero with the reason when it cannot resolve the link.

Only if that fails, fall back to CLAUDE.md → "Getting the JD": ask the user to paste the JD text into data/jd/<company>-<role>.md. Never reconstruct a JD from search snippets, and never tailor against a posting you have not actually read.

Flow

  1. Read the rules the user has already taught you, before drafting a line: python scripts/learned_rules.py brief --scope cv. These are their own corrections from earlier drafts. Re-making a correction they already made is the commonest way this skill looks like it has no memory.
  2. Parse the JD; extract must-hit keywords and the real scope behind them.
  3. Show the content sheet as plain text and get it approved before rendering anything. Headline, summary, and the bullets grouped by role, plus the cut list and why. Content, wording and layout arriving together means every correction costs a full rebuild.
  4. Build from templates/resume.html. Never template off a previous CV: a prior output carries that application's tailoring decisions along with its layout. The design system is templates/resume.css and it is linked, never pasted into the document; if a target genuinely needs a layout change, change the stylesheet and re-run bash scripts/check-resume.sh --all, so every CV moves together or none does.
  5. Draft the CV by selecting and reordering from the master source — never adding.
  6. Relevance-weighted cutting when it overflows 2 pages: score each line by (a) relevance to this posting, (b) uniqueness in the doc, (c) whether the cover letter depends on it — cut the lowest total first. An older bullet that hits posting keywords survives ahead of a recent bullet that doesn't.
  7. Spawn the Reviewer subagent (fresh context) to critique; revise, then file what it found: python gates/session.py record --event draft-reviewed --path <draft> --note "<findings>". The turn will not end until you do. If the subagent cannot be spawned in this environment, stop and tell the user — the drafter never reviews its own work as a substitute, and filing a receipt for a review that did not happen is worse than stopping.
  8. Style checker: bash scripts/check-resume.sh <draft.html> — fix everything it reports. Its last steps re-check your learned rules and the rendered page count.
  9. Honesty gate: python honesty/verify.py <draft> --target "<Company>" --job <jd-file> — must exit 0 before presenting. The flags declare the role being applied to, so naming the target company or echoing the JD's own terms is not treated as fabrication.
  10. Render and visually inspect: exactly 2 pages, no orphaned titles, fonts consistent. LaTeX is the first choice (fix layout with \needspace, \enlargethispage and re-check); if the environment has no LaTeX, fall back to HTML rendered with headless Chromium (chromium --headless --print-to-pdf=…). If no PDF renderer exists at all, present the markdown and say a PDF could not be produced here — never let a missing toolchain silently skip the visual check.
  11. Present with a verification checklist. Applying stays the human's job.
  12. If they correct the draft, run learn. One question: one-off, or from now on? A standing rule gets written to their profile in this session, not remembered by intention.

Hard rules

Every claim traces to career_facts.yaml. The gate is not optional. Scope travels with the claim: a line that is true only within a scope — one market, a pilot cohort, an average — carries that scope in the written sentence, same as it would have to be spoken. "Live in production" when it is live in one pilot market is an overclaim built entirely from real facts; the gate cannot catch it, so the drafter and Reviewer must.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.