Six skills for the documents and conversations that decide a hiring outcome, plus one we wrote for this page, measured, and deleted. How we test →
Written, reviewed and free to take. No run behind them, so no claim about what they do to an output. What that means.
If you searched for a Claude resume skill and landed on a page about the --resume flag that reopens your last session, that is a different thing entirely and this is not it. This category is about the document you send to an employer.
Almost everything written about that document is folklore. The most repeated claim on the internet is that an applicant tracking system scores your resume against the posting and rejects you below a threshold. The major systems do not work that way. What they do is parse your document into fields, store the result, and let a recruiter search and filter it, which is a completely different failure mode with completely different fixes. The thing that genuinely rejects applications automatically is the answer you type into a knockout question on the application form, and almost nobody optimises those.
We wrote a seventh skill for this category and it was the best researched file in the batch. It was about document structure: what a parser actually populates, which layout constructions leave a field empty, and the correction that the "never send a PDF" advice is folklore with no vendor documentation behind it, since the real dividing line is whether the file carries a text layer.
Then we tested it, against a control, on a case and a rubric built by an author who had never seen it. It did lift the floor: the worst run improved by exactly the margin the rule asks for. It did not clear the significance bar that the same rule attaches to that margin, so under a threshold written down before the first run it does not ship.
That is the second time this has happened on this site, and it is the same lesson both times. A skill that moved a number and still failed is the clearest evidence we can offer that the line was drawn in advance rather than fitted afterwards. The full figures are in the results table on the method page, alongside every other method we have ever run.
Its subject matter did not vanish with it. The tailoring skill below carries what a posting is actually made of, and it is explicit that document structure is out of its scope rather than quietly implying it covers it.
Resume tailoring is about what the document says: which requirements in the posting are real, which are wishlist, and how to move your own evidence to the top without inventing any of it. Its output includes a list of the requirements you cannot evidence, which is the input to the next one.
Cover letter build exists because the honest position on cover letters is contested, and the skill says so before it writes anything. Recruiter screen prep and interview story bank are the two halves of interview preparation that people usually collapse into one: the screen is a filtering conversation with a small number of specific disqualifiers, and the loop is an evidence exercise where the constraint is coverage rather than polish.
Promotion case build is the one for people who already have the job, and it is where the gap between what a strong model produces unprompted and what actually gets read in a calibration room is widest. Offer negotiation brief is last on purpose, because it is the only one where the leverage is created earlier in the process than the conversation where it gets used.
There is no LinkedIn profile skill in this category, because there is already one in the library and writing a second would be exactly the duplication this directory is supposed to avoid. Go to the profile authority rewrite for that, which carries the field caps and the truncation points as numbers.
There is no interview-answer generator, and there will not be one. A skill that writes the answer you read out in the room produces a candidate who cannot survive the second follow-up question, which is where the real assessment happens anyway.
None of the six has a number. They are published unmeasured and every card says so.
Measured means the skill was given a realistic task on real material, then the identical task was run again with the skill removed, five runs each way. Each output was graded alone, against a rubric written by someone who had never seen the skill, by a session that was not told the other arm existed. Whether it passed was decided by a rule written down before any run executed. Those pages carry the worst case, the median, the p value, and what the skill costs you as well as what it buys.
Not yet measured means exactly that. It is written, it has been read, it is free to take, and we have run no experiment on it, so we make no claim about what it does to an output. It is not a skill that failed. Skills that failed are not published at all, in either state, and their numbers are in the results table.
Measuring one skill properly costs roughly twenty model sessions. We are working down the queue and moving skills from the second group into the first. Read the full method, or go back to all skills.