resume-tailor

A resume skill with an evidence rule.

Most resume advice is about keywords. Keywords are one gate out of five, and not the one that quietly loses people. resume-tailor rebuilds a resume for one specific posting, refuses to write a claim that nothing backs, and then does the step almost nobody does: it extracts the text from the finished PDF and tells you what a parser actually receives.

The spine

No claim without evidence.

Every bullet has to trace to an entry in your evidence ledger. If it does not, it does not ship. It goes on an open-items list instead, and you get told.

This is not squeamishness. The failure mode is not getting caught, it is getting through: a resume that wins a screen on a claim you cannot defend buys you an interview you lose in ten minutes.

The five gates

A resume is read five times by three kinds of reader, in this order. Optimising for a later gate at the cost of an earlier one is the most common mistake in the genre, because if gate one fails the rest never happen.

1. Parse. A program extracts fields. Single column, standard headings, no tables or graphics, contact details in the body, static fonts.
2. Rank. Increasingly a language model scoring you against the posting. Carry their nouns, but only where your ledger backs them.
3. Scan. A human, about seven seconds, hunting six things. If those six are not findable in two seconds, nothing else on the page matters.
4. Read. Now the bullets get read. Method beats adjectives; a number beats a method.
5. Believe. A hiring manager asks whether this is real.

The detail, with sources and confidence levels on every claim, is in skills/resume-tailor/references/gates.md.

Verification, not optimisation

scripts/verify-resume.sh resume.html

Renders to PDF, then reads it back the way a parser would: page count, font embedding, whether every field survives extraction, whether section headings land on their own lines, whether a line break has welded two words together.

Put your name, phone, email, every employer, every school, and every date range in a resume.probes file beside the resume, one string per line. Every one must survive extraction. A missing probe means that field is invisible to a parser even though it looks correct on screen.

A resume you have not extracted the text from is a resume you have not checked.

What it refuses

• Hidden text of any kind, including white-on-white, near-zero font size, and instructions in metadata aimed at a language model. It is lying, it is detected at scale now, and the research shows the benefit collapses as more people do it.
• Invented or inflated numbers, including rounding a self-asserted figure up because it reads better.
• Fake titles and stretched dates.
• Keyword dump blocks.
• Chasing a third-party match score, which is a stranger's guess at what an employer runs.

Install

In Claude Code, as a plugin. This is the one to use, because it also registers the /resume-tailor-init command:

/plugin marketplace add jainulabudeenm/resume-tailor
/plugin install resume-tailor

From npm, which also covers Claude.ai and anywhere outside Claude Code:

npx resume-tailor

Add --project to install into the current repo instead of your home directory, so you can commit it and share it with a team.

Then build your evidence ledger once. On the plugin route that is /resume-tailor-init; on the npm route just ask Claude to set up your resume evidence ledger and the skill runs the same onboarding. Either way it lands at ~/.claude/resume-tailor/evidence.md, outside any repo, so it is never committed by accident.

Honesty about the evidence

Every factual claim in references/gates.md carries a tag: [measured] for peer-reviewed work or large randomised field experiments, [reported] for named industry studies with a stated method, and [directional] for vendor figures with no published method. A good deal of what circulates as resume fact is the third kind. It is labelled rather than laundered.

Two things worth knowing before you spend an evening on a document:

• A referral converts several times better than a cold application. If one is reachable, the resume is your second priority.
• Callback discrimination on names is real and measured. Nothing in this repository fixes it. It is named so that a candidate facing silence does not conclude the problem is their bullet points.

Boundaries

This clears gates. It does not produce interviews, and it emits no match percentage, because that number would be invented.

Licence

MIT. See LICENSE.