Posting in, rewritten resume out, plus a mapping document naming which requirement each change serves and which requirements are still unevidenced. It invents nothing.
/plugin marketplace add mkhalid1/locul-skills
Then /plugin to install Resume and hiring, which includes this skill.
We have not measured this one. It is published untested. What it produces is a rewritten resume plus a mapping document, and the mapping is the half that makes it checkable: one row per change, naming the posting requirement it serves, the fact in your history it rests on, and what got demoted to make room.
What it knows, concretely. First, a five-way classification of posting lines with the tells that identify each class from the text alone, including the one most people miss: a hard gate is not the line in the posting, it is the Yes/No question on the application form. Greenhouse's own documentation for its auto-reject rule says the rule fires on the answer to a single application question, and that the people who normally get notified about new applications are not notified about auto-rejected ones. Verified on Greenhouse's support site, August 2026.
Second, evidence that the wishlist class is real rather than a comforting story. In the Harvard Business School and Accenture report *Hidden Workers: Untapped Talent* (3 September 2021), based on a survey of 2,275 executives across the United States, the United Kingdom and Germany, 88 per cent of employers agreed that qualified high-skills candidates get vetted out for not matching the exact criteria in the job description, and only 21 per cent said all of their high-skills hires over the previous three years had met every requirement listed in the posting.
Third, a substitution rule stated as a swap. Every promotion demotes something, because a resume that grows during tailoring is a resume that got worse.
Who it is not for. If your problem is that the file arrives unreadable, this fixes nothing. If you are applying to sixty postings a week, the per-posting cost here is wrong for you and a scanner is the better tool.
Measuring one skill honestly costs about twenty model sessions: five runs with it, five without, on real material, each output graded alone by a session that is not told the other arm exists, against a rubric written by somebody who never saw the skill. We have not spent that on this one yet, so it ships labelled rather than ships silently.
How it would be measured. Tier A. Material: twelve invented job postings written by an author who never saw this file, each line labelled in advance as hard gate, load bearing, screening vocabulary, wishlist or boilerplate, paired with six invented resumes whose facts are fixed and enumerable, five runs per arm, graded blind against the pre-recorded labels. Objective spine: correct class per posting line, every promoted bullet traceable to a fact present in the source resume, resume length not increased, each promotion paired with a named demotion, the unevidenced list present and complete, and zero claims in the output that do not appear in the input.
The spine is unusually clean for something in this category. Posting lines can be labelled in advance by an author who never saw the file, so classification accuracy is countable. So is the fabrication check, which matters more: every claim in the output either traces to a fact in the input resume or it does not, and that is a yes or a no per bullet.
The awkward part is the material. Real postings cannot be used, because a public one carries an employer's name, and invented postings written by somebody who knows the taxonomy drift towards containing the tells on purpose. The honest prior is that a strong model already produces a plausible tailored resume unprompted. The narrower questions are whether it holds the length constant, whether it names the demotion, and whether it refuses the wishlist items instead of quietly claiming them.
The rule that decides pass or fail was written down before any run was executed and it does not move afterwards. It is in the method note on the hub, along with the full results table including every skill that was tested and cut.
Stated plainly, because a skill that claims everything is useful for nothing.
name and description in the file's frontmatter, so you can also invoke it by name.A recruiter in your field, given twenty minutes and the posting, does the classification faster and better than any file, because they know which line the hiring manager wrote. If you have access to one, that is the answer.
Scanners are the honest commercial alternative and they do a different job well. They compute term overlap, which is a real signal for retrieval, and they see the posting's vocabulary without the judgement layer this file adds. What they cannot do is tell you that seven of the eleven listed tools are wishlist padding, or stop you writing a skills block that a human reads as a lie.
The model with no skill at all writes a decent tailored resume on the first try. The things it tends not to do are the unglamorous ones: refusing to add a tool you have never touched, keeping the page count fixed, saying out loud which bullet it pushed down the page, and producing the list of requirements you cannot evidence, which is the part that decides whether the application is worth sending at all.
--- name: resume-tailoring-pass description: Rewrites an existing resume against one named job posting by first sorting every line of that posting into hard gates, load bearing duties, screening vocabulary, wishlist and boilerplate, then promoting real evidence by swap rather than by addition. Produces a rewritten resume plus a mapping document recording which requirement each change serves, what was demoted to make room, and which requirements remain unevidenced. Carries the tells that identify each requirement class from the text alone, the bullet form, the length rule, and an apply-or-not decision. This skill should be used when a resume already exists and has to be aimed at one specific posting before the application goes in. --- # Resume tailoring pass ## What this produces, and why the obvious approach fails Two files come out of this. A rewritten resume, and a mapping document: one row per change, naming the posting requirement it serves, the fact in your history it rests on, and what was pushed down the page to make room. The mapping is not optional. It is the only thing that lets you, or anybody else, check that the rewrite added no experience you do not have. The obvious approach is to paste the posting in and ask for the keywords. It produces a resume that mirrors the posting's vocabulary and says nothing the posting did not already say, and the reason is structural rather than a failure of effort. A job posting is not written by one person. It is assembled by three, with three different jobs. A hiring manager writes what the work actually is, usually in the first few responsibilities and usually in plain verbs. A recruiter adds the terms they will type into the database search box, which is why the tool list is longer than any one person's job. And a template supplies the rest: the equal opportunity statement, the benefits, the culture prose, other duties as assigned, and in a growing number of jurisdictions a legally required pay range. Mirroring all three equally is why tailored resumes read as generic. The recruiter's terms are for retrieval, not judgement. The template's text carries no requirement at all. Only the hiring manager's part describes the job, and it is usually the shortest part of the document. The requirement accretion is documented. In *Hidden Workers: Untapped Talent*, published by Harvard Business School's Project on Managing the Future of Work with Accenture on 3 September 2021, from a survey of 2,275 executives across the United States, the United Kingdom and Germany conducted in January and February 2020, 72 per cent of employers said that when creating a new posting for a middle-skills role they used the existing posting or slightly modified it, and only 8 per cent wrote a new one. The report describes the result as postings larded with legacy requirements and nice-to-have attributes rather than a limited list of must-have skills. Requirements are added; they are almost never removed. All figures in this file were verified against that report in August 2026. ## Step 1. Classify every line of the posting Do this before you open the resume. Copy the posting into a working file and label every line with one of five classes. Nothing is unlabelled, including the sentences that look like scenery, because deciding they are scenery is the label. **HARD GATE.** Something the application genuinely stops on: work authorisation or sponsorship status, a professional licence, a security clearance level, a named certification with an issuing body, a legally required qualification, sometimes a location or an on-site requirement. Tells: the line uses must, required, or is a condition of employment. It names an instrument that has an issuer and can be checked. And critically, it reappears as a question on the application form. That form question is where the rejection actually happens. Greenhouse documents an auto-reject application rule that fires on the answer to a single Yes/No, single-select or multi-select application question, rejects the candidate automatically, and does not notify the staff who normally receive new-application alerts, so nobody reads the resume at all. Verified on Greenhouse's support documentation, August 2026. Other systems ship equivalent features under names like knockout questions. The practical consequence: no phrasing on a resume defeats a hard gate. If you meet it, state it plainly and early in unambiguous words. If you do not, no tailoring changes the outcome through the form. **LOAD BEARING.** What the role does most days. Usually two to four items. Tells: it appears in both the posting's summary paragraph and the responsibilities block. It sits in the first three responsibilities. It is written with a direct object and a frequency, such as owning a weekly forecast across forty sites, rather than as an adjective. And it survives the deletion test: if you struck out everything else in the posting, this is still recognisably the job in the title. **SCREENING VOCABULARY.** Named tools, platforms, frameworks and methodologies. Tells: proper nouns, often with vendor names or version numbers, appearing in a comma-separated run rather than a sentence, and often repeated in the posting's skills chips. These matter for retrieval and not for judgement. A recruiter types them into a search box. In the same report, 63 per cent of surveyed employers used a recruiting management system, rising to 75 per cent in the United States, and more than 90 per cent of those used it to filter or rank candidates in the initial screen. So the terms are worth placing accurately, exactly once each, in the role where they are true. **WISHLIST.** The nice-to-have block, and anything sitting inside a long list. Tells: it is under a heading reading preferred, bonus, plus, or nice to have. Or it sits inside a bulleted list of nine or more items. Or the combination asked for would take longer to acquire than the years of experience the posting requests, which is the arithmetic tell and the fastest one. That these are aspirational is measurable from the employer side. In the same survey, 88 per cent of employers agreed that qualified high-skills candidates are vetted out for not matching the exact criteria in the job description, rising to 94 per cent for middle-skills roles. And when asked about people they actually hired, only 21 per cent of employers said all their high-skills hires over the previous three years had met every requirement listed in the posting, while 47 per cent said half or fewer of their middle-skills hires had. A note on the folklore here, because it is repeated constantly. The claim that men apply at 60 per cent of the listed qualifications and women only at 100 per cent appears in that same report, and its footnote points to a single 2014 Harvard Business Review essay titled *Why Women Don't Apply for Jobs Unless They're 100% Qualified*, which opens by presenting the number as one the reader has probably heard already. The chain of citation ends in received wisdom rather than in a study. Treat the underlying point, that long requirement lists suppress applications, as supported by the employer survey above, and treat the specific percentages as unsourced. Verified August 2026. **BOILERPLATE.** Equal opportunity statements, other duties as assigned, benefits, values prose, the application process description. Tells: the identical wording appears in other current postings from the same employer, which takes two minutes to check and is the most reliable tell in this file. Also legally mandated content, which is not a signal about the role at all: Colorado's Equal Pay for Equal Work Act, SB19-085, effective 1 January 2021, requires employers to announce job openings to employees together with the pay range for those openings, and New York Labor Law section 194-b requires employers with four or more employees to state a compensation range in the advertisement. A salary line in a posting from either jurisdiction tells you about the statute, not about the role's seniority. Boilerplate is the single biggest source of noise when a model is asked to match a posting, because it is often the majority of the word count and it is written in confident, requirement-shaped English. ## The decision rule for a badly written posting Most postings are badly written, so this is the branch you will use rather than an edge case. - **The line appears in both the summary paragraph and the responsibilities block.** Load bearing. Classify it there even if it also names a tool. - **The line appears only inside a bulleted list of nine or more items.** Wishlist. The length of the list is doing the work, not the wording. - **The line names an instrument with an issuer and the posting says must.** Hard gate. Then go and open the application form and look for the matching question before you write anything. - **You cannot tell.** This happens most often with a methodology named once, in a sentence that could be a duty or could be a preference. Treat it as screening vocabulary. Place the term once, accurately, in the role where it is genuinely true, and restructure nothing around it. Do not promote a bullet for it, do not put it in the summary, and do not add it to a skills block if it is not true anywhere. The cost of a misfiled screening term is one word in the wrong place. The cost of restructuring a resume around a wishlist item is a page that argues for a job nobody is hiring for. ## Step 2. Build the evidence inventory before you rewrite anything Now open the resume, and any other source material you have: performance reviews, project write-ups, old versions of the file, notes. Build a table, and build it from the source material only. One row per thing you have actually done, with four columns: what you did, the outcome, the number, and where in your history it sits. Rows with no number stay in the table with the number column empty, because an empty cell is information and a guessed number is not. **The rule that governs everything after this point: nothing enters the resume that was not already true in the source material.** This skill rewrites, reorders and quantifies. It does not invent. That includes the softened forms, which is where invention actually happens in practice: exposure to, familiarity with, worked alongside, supported the team that. If the underlying fact is that you sat near somebody who used the tool, none of those phrasings make it true, and every one of them is a question you will be asked in the screen. The inventory is also where you discover you have evidence you had forgotten, which is the common case and the reason this step precedes the rewrite rather than following it. A person who reads the posting first sees their history through the posting's vocabulary and stops noticing anything the posting did not name. ## Step 3. The substitution rule For each load bearing requirement, find the strongest real piece of evidence and promote it. There are exactly two promotion targets: the first bullet of the most recent role where it is true, and the summary. Nothing else on a resume gets read reliably enough to count as promotion. **Every promotion is a swap, not an addition.** When a bullet moves up, another moves down or comes out. Record both in the mapping document. The reason is not aesthetic. A resume that gains half a page during tailoring has diluted every strong line on it, and the person reading it spends their attention budget on the material you added rather than the material you promoted. Fix the length before you start, in lines or in pages, and hold it. **The bullet form: outcome first, then the action, then the scale or constraint.** Outcome first because the first six or seven words are the ones that survive a skim. Scale last because it qualifies the claim rather than making it. A duty bullet and an outcome bullet on the same underlying work, from an invented example: > Duty form: Responsible for managing the monthly close process and maintaining reporting for the operations team. > Outcome form: Cut the monthly close from eleven working days to six by rebuilding the reconciliation checklist, across four entities and two currencies. Same work. The first says what you were assigned. The second says what changed, and it is defensible in a conversation because every element of it is checkable by the person who asks. Three constraints on the rewritten bullet: 1. **The number must come from the inventory.** If the inventory cell was empty, the bullet is written without a number. A bullet with a plausible invented number is the worst artefact this process can produce, because it survives the resume screen and fails in the room. 2. **Do not inflate the verb.** Led means you led. Contributed to means you contributed. If your source material says you were one of four people on a project, the rewrite says so or says nothing about team size. 3. **One requirement per bullet.** A bullet that serves three requirements at once serves none of them legibly, and it cannot be mapped. ## Step 4. The unevidenced list, and the apply-or-not call Every requirement with no supporting row in the inventory goes on a list. **This list is an output, not a failure.** It is what a [cover letter](/skills/cover-letter-build/) addresses, what a screen call prepares for, and what tells you whether to send the application at all. Sort it by class, because the classes carry different consequences. An unevidenced hard gate is decisive. An unevidenced load bearing requirement is the real question. An unevidenced wishlist item is, on the evidence above, usually nothing. The apply-or-not rule, which does not pretend to certainty because you cannot see the other applicants: - **Any hard gate unmet, and it reappears as a form question.** Do not apply through the form. The answer decides it before a person is involved. The only route left is a person who can route around the form, and that is a different piece of work. - **All hard gates met, and two or more load bearing requirements have real evidence with outcomes you could defend in a fifteen-minute conversation.** Apply. The unevidenced list becomes cover letter and screen preparation material. - **All hard gates met, exactly one load bearing requirement evidenced.** This is a stretch application and should be named as one. Send it if the cost is low or if you can name a person to send it to. If neither is true, the same hour spent on a posting where two are evidenced is worth more. - **No load bearing requirement evidenced.** Tailoring will not fix this. The document is not the problem. ## Step 5. Write the mapping document Short, tabular, and delivered alongside the resume: | Posting line | Class | Evidence row used | Change made | Demoted to make room | | --- | --- | --- | --- | --- | Then three closing blocks: the unevidenced list sorted by class, the length check as a before and after figure, and a one-line fabrication statement confirming every claim in the rewrite traces to a row in the inventory. If you cannot write that last line honestly, the rewrite is not finished. ## Worked example A mid-size logistics company posts for a Supply Chain Data Analyst. The candidate is a business analyst at a billing service. Everything here is invented. **Three requirements classified.** 1. *"Eligible to work in the country without sponsorship, now and in the future."* Hard gate. The application form carries a matching Yes/No question. The candidate meets it, so the resume states it once, plainly, near the top. 2. *"Own the weekly demand forecast for 40 distribution sites and present variance to the operations director."* Load bearing. It appears in the summary paragraph and again as the first responsibility, it has a direct object and a frequency, and it survives the deletion test. 3. *"SQL, Power BI, dbt, Snowflake, Airflow, Python, R, Tableau, Looker, Excel, VBA."* Eleven items in one comma-separated run. Screening vocabulary. The candidate has used four of them. Four go in, in the roles where they are true. Seven do not go anywhere. Also present and labelled: a nine-item preferred block containing sales and operations planning process design, which is wishlist, and roughly 40 per cent of the word count in benefits, values prose, other duties as assigned and an equal opportunity statement, which is boilerplate. Two of the boilerplate paragraphs appear word for word in two other current postings from the same employer. **Two substitutions.** Summary, before: *Detail-oriented business analyst with six years of experience across finance and operations, seeking to apply analytical skills in a fast-paced environment.* Summary, after: *Business analyst who owns a weekly demand forecast across 18 client accounts and cut forecast error from 22 to 13 per cent over four quarters.* Most recent role, first bullet, before: *Responsible for building and maintaining reports for the operations team.* After: *Cut weekly forecast error from 22 to 13 per cent over four quarters by rebuilding the demand model in SQL and rebasing it on shipped units rather than invoiced units, across 18 client accounts.* Demoted to make room: a bullet about migrating the reporting stack moves from first position to fourth. It is true, it is good, and it serves no load bearing requirement in this posting. Length before, two pages. Length after, two pages. **One requirement left unevidenced.** Sales and operations planning process design. The candidate has participated in the meeting and has never designed the process. Nothing is added to the resume. The row goes on the unevidenced list, marked wishlist, with a note that the adjacent true fact is four quarters of forecast ownership feeding that meeting. **Verdict.** Hard gate met and stated. Two of the three load bearing requirements carry evidence with numbers that came from the inventory rather than from the posting. Seven tool names were declined. One wishlist item is unevidenced and stays unevidenced. Apply, with the sales and operations planning gap as the first thing the cover letter deals with and the first thing to rehearse for the screen. ## Failure modes **Mirror drafting.** The rewrite becomes a paraphrase of the posting. Every requirement is echoed, in the posting's own words, and the document reads as nobody in particular. The tell is that you could reconstruct most of the posting from the resume. **Keyword ballast.** A skills block containing terms that appear nowhere in the work history. It fails twice: a human reads an unsupported list as a lie, and a recruiter's search returns you for a role you cannot do, which produces a screen call that ends badly rather than no call at all. **Silent promotion.** A claim inflated during rewriting so the candidate cannot defend it. Contributed becomes led, a team result becomes a personal one, a pilot becomes a rollout. It survives the screen and dies in the room, where the follow-up question is always what did you personally do. **Length creep.** Every promotion is added and nothing is removed, so the file grows by a third. It looks like more evidence. It reads as less, because the strongest line is now surrounded by four merely adequate ones. **Recency inversion.** An older but on-topic role is moved above a recent one to put the matching work first. Chronology is the one structural expectation a reader has, and breaking it reads as career regression rather than as emphasis. Promote the bullet, never the role. **Boilerplate chasing.** The rewrite responds to the values prose and the benefits section, producing summary lines about thriving in collaborative fast-paced environments. This is the failure mode a model falls into most readily, because boilerplate is often the majority of the posting's word count and is written in requirement-shaped sentences. **One resume, many jobs.** The file is tailored hard to a single posting and then sent to six others, where the summary now names a specialisation the reader did not ask for. Tailoring is per posting or it is worse than not tailoring. Keep a base version and never overwrite it. **Gate blindness.** Hours spent on wording for a posting whose form asks a Yes/No question the candidate answers wrongly. The document was never read. Check the form first. ## What this skill does not do - It does not fix whether the document parses. Layout, section headings, date formats and file format are a separate job and out of scope here: this file only changes what the document says. A resume scanner that uploads your file to a real parser and shows you the extracted record will tell you more about parsing than any prose can. - It cannot see the applicant tracking system, the recruiter's saved search, the scoring configuration if one exists, or the other applicants. It reasons from the posting text and the form questions, which is what you can actually see. - It does not decide whether the job is worth having. Compensation, the manager, the team and the trajectory are outside it entirely, and it will happily help you aim carefully at something you should not want. - It will not invent experience, in any form, including the softened ones. That is a refusal rather than a limitation, and it is the reason the mapping document exists. - It is not a substitute for a referral. A warm introduction changes how the same document is read, and nothing in this file matches one.
These skills all ask your assistant to check things against your actual codebase, your actual schema, your actual design system. Locul keeps that context current on its own, from the files you already have, on your machine. Mac and Windows, free to start.