Skills/Resume and hiring/Cover letter build

Cover letter build: decide whether it will be read before deciding what it says

Posting in, one letter out, plus a line naming the reader it was written for. Five arrival situations, each with its own length, opening move and cut list.

Not yet measured skill 3,430 words MIT by Locul Verified safe · 0 secrets Written 2026-08-20
/plugin marketplace add mkhalid1/locul-skills

Then /plugin to install Resume and hiring, which includes this skill.

Download just this file3,430 words
No account and nothing to sign up for. The marketplace command works in Claude Code today. Using something else? Copy the file or download it and put it where your assistant expects.
We have not measured this skill. There is no result on this page because we have not run one. It is written, it has been read for accuracy, and it is free to take. Nothing below claims it improves an output, because we have not shown that. This is different from a skill that failed our test: those are not published at all.
What it is, and what we are not claiming

Untested. The asset is the five-way classification of how a letter arrives, and the branch where the honest output is no letter at all.

We have not measured this one. It is published untested, and the first thing it does is refuse the question everyone asks. Whether cover letters are read is genuinely contested and the evidence is thinner than either side claims, so the skill prints what the sources actually support, with samples and dates, and then moves to a question that is answerable: whether this particular letter, arriving this particular way, will be read.

What it knows, concretely. First, the evidence base itself, named rather than gestured at. One field experiment exists, published by ResumeGo in 2020, covering 7,287 fictitious applications submitted between 15 July 2019 and 10 January 2020, which reported a 53 per cent higher callback rate for tailored letters than for none. Everything else is self-reported surveys, and the reading rate they produce ranges from about 60 per cent to 87 per cent depending on who ran the survey, all of them companies selling resume or cover letter products.

Second, a five-way classification of how a letter arrives, because the arrival decides the length before any content is chosen. A required text box on a form takes 150 to 200 words. A direct email to a named person is not a letter at all and takes 120 to 180. A referral introduction takes 100 to 150 and has to be forwardable. A visible question on the resume takes 200 to 300. An optional upload on a high-volume portal with nothing to explain takes zero, and the skill says so.

Third, a slot structure with word budgets and one ratio rule: the objection never gets more words than the evidence, which is what stops a gap turning into three defensive paragraphs.

Who it is not for. If you have a referral, the letter matters much less and this is a rounding error on your odds. If you are applying to fifty postings this week, the honest answer is that this file will tell you not to write most of them.

When to reach for it

  • At the moment the application form loads and the cover letter box turns out to be marked required, which is the point a letter stops being optional and starts being a disqualifier.
  • Immediately after a resume has been tailored to a posting, while the list of requirements that resume cannot evidence is still on screen, because that list is the only raw material a letter has.
  • When somebody has offered a referral and asked for something they can forward, which is a shorter and differently shaped document from the one the portal wants.
  • Before applying to a role that sits across a career switch, a gap of more than a few months, or a relocation, where the reader forms the question whether or not you answer it.
  • When the same letter is about to be lightly edited for a twelfth application, which is the point the previous company's name gets left in the second paragraph.

Why there is no number on this page

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 B with a Tier A spine. Material: twenty invented postings whose application context was fixed in advance by an author who never saw the file, ten with a required cover letter field, five optional on a high-volume portal, five arriving as a direct email to a named person, half of them carrying a visible resume question such as a gap, a switch or a relocation, each paired with an invented candidate history and a pre-written list of posting requirements the resume cannot evidence. Objective spine: correct situation classification per posting, word count inside the band for that situation, at most three unevidenced requirements carried, an objection sentence present where a visible question exists and absent where it does not, no opening drawn from the banned list, a named addressee used where the posting contains one, and a no-letter verdict returned on the optional-portal cases with nothing to explain. Prose quality on top of that would need blind pairwise preference.

The spine here is unusually clean for something that ends in prose. Situation classification, word count against the band for that situation, the number of unevidenced requirements carried, the presence or absence of an objection sentence, and whether the opening appears on the banned list are all countable by somebody who never saw the file.

The awkward part is the material. Application context has to be written into the invented postings deliberately, since the classification turns on whether the field is required, whether a name is discoverable and whether the resume carries a visible question, and none of that is visible in posting text alone.

The honest prior is that a strong model writes a competent, conventional cover letter unprompted. The narrower questions are whether it returns "do not send one" on the optional-portal case, whether it holds the objection to a single sentence, and whether it stops at three requirements.

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.

What it does not do

Stated plainly, because a skill that claims everything is useful for nothing.

  • It cannot tell you whether one specific employer reads cover letters. No published source breaks reading rates down by company, so the classification here is inference from how the letter arrives, not knowledge of who opens it.
  • It cannot make a weak application strong. If the resume does not already evidence the top requirements, a letter arguing around three of them is arguing from the wrong document.
  • It never touches the resume, the file format or the application form, so it will happily produce a good letter attached to a document that arrived in the system with an empty work history field.
  • It will not write an emotional appeal. There are hiring situations, small organisations and mission-led teams among them, where one lands better than anything this file produces.
  • The plain language guidelines published at plainlanguage.gov beat it on sentence-level prose, and a person who already works there beats the whole file, because a referral moves an application further than any letter.
  • It stops at send. Follow-up timing, chasing a silent application, and what to write after a rejection are all out of scope.

Install it

  1. Open Locul, go to Library, and choose Import. One-click import from this page lands shortly.
  2. Locul writes the file to the right folder for every assistant you have connected, so you do not have to know where each one keeps its skills.
  3. Environment variables and headers in any shared config are replaced with a placeholder before they reach you, so importing a stranger's setup cannot hand you their credentials or take yours.
  4. Locul is free to start, on Mac and Windows. Get it here.
  1. Download SKILL.md using the button above, or copy the file.
  2. Save it at .claude/skills/cover-letter-build/SKILL.md in your project, or under ~/.claude/skills/cover-letter-build/SKILL.md on Mac and Linux, or %USERPROFILE%\.claude\skills\cover-letter-build\SKILL.md on Windows, to make it available everywhere.
  3. Start a new session. Claude Code picks up the skill from the name and description in the file's frontmatter, so you can also invoke it by name.
  1. Download or copy the file.
  2. For Claude Desktop, add it through the skills panel in settings, or drop the folder into your skills directory.
  3. For Cursor and other assistants that read plain instruction files, paste the body into your project rules file. The skill is plain markdown with no tool bindings, so it carries across.

Pairs well with

What else does this job

Writing it yourself in fifteen minutes is often the right answer, and it is the one this page would rather you took when the situation is a required text box and the letter exists to avoid an empty field. Nothing here is hard. The taxonomy exists because the length decision is made badly under time pressure, not because the sentences are difficult.

The published guides are good and free. The plain language guidelines are better than this file at the prose. A university writing lab gives you a conventional model to copy, which is exactly right when the reader is expecting a conventional letter and nothing about your application is unusual.

The model with no skill at all produces a fluent, well-organised letter on the first try. What it tends not to do is refuse. It will write 400 words for an optional upload where the correct output is no letter, it will spend a paragraph apologising for a gap that deserved one sentence, and it will open with a line that roughly a third of the pile now opens with.

Read the full source
---
name: cover-letter-build
description: Produces one cover letter, plus a single line naming which reader it was written for and why. Classifies how the letter will arrive into five situations, each with its own word budget, opening move and cut list, and returns a no-letter verdict where the expected value is near zero. Carries the published evidence on whether letters are read with samples and dates, a five-slot structure with per-slot word counts, an ordered rule for finding an addressee, and eight opening sentences that mark a letter as templated. This skill should be used when someone is applying for a job and has to decide whether to write a cover letter and what it should say.
---

# Cover letter build

## What the evidence actually says, and why it does not settle anything

Start here, because the whole method follows from the state of the evidence rather than from a position about it.

**One field experiment exists.** ResumeGo, a resume writing company, published a study in 2020 covering 7,287 fictitious job applications submitted between 15 July 2019 and 10 January 2020 to postings on three large job boards. Three arms: no cover letter, a generic one, and one tailored to the posting. The reported result is that tailored letters drew a 53 per cent higher callback rate within 30 days than sending none.

Take it seriously and take it with the caveats. The publisher sells the service the study finds valuable. Only the relative difference was published, never the underlying callback rates, so a rise from a small number to a slightly larger small number looks the same as a large one. The outcome measured is a callback, not an offer. The applications were fabricated, and nobody without a commercial stake has repeated the study.

**Everything else is people saying what they do.** ResumeLab, 2019, 200 hiring managers: 74 per cent read the letter when the posting required one. ResumeGo's own survey alongside the experiment, 236 hiring professionals: 87 per cent read them. GetCoverLetter, 2020, 2,000 recruiters: 78 per cent favour applicants who send one. Resume Genius, 2023, 625 US hiring managers: 83 per cent read most cover letters and 45 per cent read the letter before the resume. Novorésumé's hiring landscape survey, 2025, more than 200 HR professionals: about 60 per cent read them, 39.4 per cent do not read them at all, and recruiters specifically are the least likely, with 52.8 per cent skipping them entirely.

**The spread is the finding.** The same nominal question returns 60 to 87 per cent across five surveys in six years, and every publisher sells resume or cover letter products. Question wording and panel composition move that number more than any plausible change in employer behaviour. "Read" is never defined and covers a three-second glance. Asking a hiring professional whether they read the letters candidates send them is asking whether they do their job properly, which is textbook social desirability pressure. Samples in the low hundreds cannot be split by industry, seniority or application volume, and that is precisely where the variation lives.

**One signal survives the noise, and the method is built on it.** The 2025 split between recruiters and hiring managers points the same way in every source that reports both: the person working a queue of applications skips, and the person who owns the vacancy reads. That is a fact about who receives the letter rather than about letters, and who receives it is something you can usually work out.

**Two current numbers about the pile your letter joins.** Novorésumé surveyed 1,000 US job seekers aged 18 to 64 in July 2026: 46.8 per cent used AI to some extent on their last cover letter and 32.4 per cent have submitted one written almost entirely by AI with little or no editing at least once. Separately, Insight Global's 2025 AI in Hiring survey of 1,005 US hiring managers found 88 per cent said they can tell when candidates use AI. Treat that as a self-assessment of detection rather than a measurement of it. The operational consequence is not that anyone catches you. It is that the standard openings are now the modal opening in the stack.

## The five situations, which set the length before you write a word

Classify first. The length and the opening move come out of the classification, not out of taste.

**1. Required field on the application form, with a text box.** The reader is a recruiter or a coordinator working a queue. Skimmed rather than read, and skimmed for disqualifiers: wrong role, wrong location, wrong level, no evidence of the one hard requirement. **150 to 200 words.** Open with the role name and the single strongest matching fact in the same sentence. Cut every sentence that restates the resume, all narrative, and every mention of what you are looking for in your next role. The job here is to be unobjectionable and specific, and an empty required box is the only outcome that actually costs you anything.

**2. Attached to a direct email to a named person.** Read almost always, because a person chose to open it. This is not a cover letter, it is an email, and treating it as a letter is the error. **120 to 180 words in the body**, no attached document unless one was asked for. The subject line does half the work: role, and the one fact that makes you plausible. Cut "please find attached", cut the formal salutation ceremony, cut anything that sounds like a document rather than a message.

**3. A referral introduction.** Read, but the referral is doing the work and your letter is not the persuasion, it is the thing your contact forwards. **100 to 150 words.** The referrer's name goes in the first six words. Cut the persuasion entirely and keep it short enough to paste into somebody else's message without editing.

**4. A gap, a switch, a relocation, or an obvious mismatch with the posting.** Read, because the reader already has a question and the letter is the only place an answer could be. This is the situation where a letter earns its place regardless of the survey numbers. **200 to 300 words.** Open with the hook, answer the question in slot four, and cut every word of apology.

**5. Optional upload on a high-volume portal, with nothing on the resume to explain.** Effectively unread. The reader is the queue worker from situation one, and the field was optional because somebody configured it that way. **Zero words.** Write no letter and put the time into the resume, which is read in all five situations.

## The input: the list of things your resume cannot evidence

This skill does not start from a blank page. It starts from the output of the [resume tailoring pass](/skills/resume-tailoring-pass/): the list of requirements in the posting that the resume, as written, does not evidence.

Sort that list by whether a human can be persuaded.

- **A hard credential you do not hold.** A licence, a clearance, a degree that is a legal requirement. Not persuadable, and raising it wastes words. Ignore it unless a genuine equivalence exists, in which case it gets one clause.
- **Years of experience slightly short.** Weakly persuadable, and never by arguing about duration. Answer with a scope fact: what you were responsible for, at what size, with what constraint.
- **A tool or stack you have not used, where you have used an adjacent one.** Persuadable, and the most reliably persuadable item on the list. Name the adjacency and the specific transfer.
- **A domain you have not worked in.** Persuadable if you can state the domain's central constraint in one sentence in a way that shows you understand it.
- **Scope or seniority.** Persuadable with a number that has a denominator attached.

Then take at most three, and carry two or three into the letter. Never more. Anything the resume could evidence goes on the resume instead, because a claim on the resume reads as a fact and the same claim in the letter reads as an assertion.

## Structure: five slots with word budgets

Budgets given for the 200 to 300 word version. Scale proportionally for the shorter situations, and drop slot four entirely when there is no visible question.

**Slot 1, the hook. 25 to 40 words.** One specific, checkable fact about the company, the product, the role or the market, of the kind only available to somebody who read documentation, used the product, or followed the industry. The test: could this sentence be pasted unchanged into a letter to their nearest competitor? If yes, it is not a hook, it is flattery.

**Slot 2, the claim. One sentence, 20 to 30 words.** The sentence a reader could repeat to a colleague without looking back at the page. Shape: what you are, plus the thing you have done that this job needs. If the sentence cannot survive being said out loud in a corridor, it is not the claim.

**Slot 3, the evidence. Two items, 40 to 70 words each.** At least one carries a number with its denominator stated. These are not resume bullets in prose. The resume says what you did. The letter says how, and what it cost, and what you decided.

**Slot 4, the objection. 25 to 45 words, one sentence, two at the absolute limit.** State the fact in the first clause and answer it in the second. Then stop. **The ratio rule: the objection never gets more words than the evidence.** A gap that occupies more of the page than your work is a letter about a gap.

**Slot 5, the close. 15 to 25 words.** An action with a subject and a verb. Not a hope, not a thank you, not "I look forward to hearing from you".

The order is not arbitrary. The hook buys the second sentence. The claim is what survives if the reader stops there, which is the likeliest outcome. The evidence exists to make the claim repeatable by somebody arguing for you in a room you are not in. The objection comes fourth because answering it first makes the whole letter about the problem. The close is last because it is the only line that is about them.

## The addressing problem

Work down this list and stop at the first hit.

1. The posting itself. A hiring manager's name appears in the posting more often than applicants check.
2. The application email address, and the sender name on any confirmation email you have already received.
3. The company's own site: a team page, an engineering or company blog, a documentation changelog, a press release naming the department head.
4. The company's public profile pages, filtered to the department rather than to recruiters.

If you find a name, address it as "Dear" plus the full name. Do not add a title, because titles guess at gender and at marital status, and both guesses are avoidable. Use a stated honorific only where the person publishes one.

If you find no name: "Dear Hiring Team," or the department, as in "Dear Data Platform Team,". Both are fine and neither is a flaw.

Never use "To Whom It May Concern", which reads as a document you did not write for them. Never use "Dear Sir or Madam". "Dear Hiring Manager" is tolerable and generic, and it is what you fall back to when you do not even know the team name.

**The cannot-tell branch.** If two plausible names exist and you cannot establish which one owns the role, use the team form. A letter addressed confidently to the wrong person is worse than one addressed to nobody, because it is evidence you guessed.

## Openings that mark a letter as templated

Each of these signals something specific to a reader who has seen several hundred.

- "I am writing to apply for the position of X at Y." A template with two fields filled in, and it spends the one sentence guaranteed to be read on information already in the form.
- "I am excited to apply" and "I was thrilled to see this role". Enthusiasm asserted before anything has been shown.
- "As a passionate <role> with N years of experience". The word passionate is doing work that a verb in the past tense should be doing.
- "Your commitment to <sentence lifted from the About page> deeply resonates with me." Mission mirroring. It proves you opened one page.
- "When I saw this role, I knew I had to apply." A feeling, and nothing about it can be false.
- "I believe I would be a great fit for this role." States the conclusion the rest of the letter was supposed to earn.
- "In today's rapidly changing <industry> landscape." An essay opening on a document that is not an essay.
- "Allow me to introduce myself." Ceremony, and 25 words of it.

This list matters more now than it did in 2019. With roughly a third of job seekers having sent a near-unedited AI letter at least once, as of the July 2026 survey above, these are no longer merely weak openings. They are the openings the reader has already seen twice on this screen.

## The decision rule: whether to write one at all

- **The form marks the field required, or the posting says to include one.** Write the situation one version, 150 to 200 words. Not a judgement call. An empty required field is a disqualifier you can avoid for twenty minutes of work.
- **You have a named human and a direct email address.** Write the situation two version. This has the highest read probability of the five and the lowest word count of the written options.
- **You have a referral.** Write the situation three version, forwardable, and keep it short enough to paste.
- **Optional on a portal, and the resume already evidences the top three requirements.** Do not write one. The expected value is near zero and the same hour spent on the resume is read in every situation.
- **Optional on a portal, but the resume carries a visible question.** Write the situation four version. This is the branch most people get backwards: the letter earns its place because of what is missing from the resume, not because of what is in the posting.
- **You cannot tell.** The posting is on an aggregator that stripped the original form, you cannot see whether the field is required, or you do not know who receives it. **Write the short version, 150 to 200 words.** The cost is bounded at about twenty minutes and the downside of an empty required box is worse than the downside of an unread short letter. Do not write the long version here. The cannot-tell case is exactly where an afternoon disappears.

## Worked example

An analyst with six years in retail supply chain planning applies for "Senior Analyst, Marketplace Operations" at an invented mid-size online marketplace. The posting is on the company's own careers page. The cover letter field is present and marked optional. No name appears anywhere in the posting.

Requirements the resume evidences: SQL, demand forecasting, stakeholder reporting. Requirements it cannot evidence: three or more years in a marketplace or two-sided platform business, experience partnering with category managers, and one named BI tool the candidate has never opened.

**Classification.** Optional field, so situation five is the default. But the resume shows six years in retail planning against a marketplace operations role, which is a visible switch, and the reader will form the question. Situation four applies. Length band 200 to 300 words.

**Addressing.** No name in the posting. The careers page names a Marketplace Operations team but no head. Two plausible names appear on the company's public department pages with no way to establish which owns the role, so the cannot-tell branch applies: "Dear Marketplace Operations Team,".

**Unevidenced list, ranked.** Marketplace domain, persuadable, and it is the same thing as the visible switch, so it becomes slot four rather than a separate argument. Category manager partnering, persuadable, becomes one of the two evidence items. The BI tool, weakly persuadable by adjacency, gets one clause inside an evidence sentence and no more.

**Slot four, as written.** "I have not worked inside a marketplace, so the supply side of one is new to me. What is not new is forecasting demand for an assortment I do not control, which is what six years of planning against third-party vendors was." One sentence plus one, 46 words, no apology, and the fact is stated before the answer.

**Verdict.** Send it, and the reason is the switch, not the posting. Had the same candidate applied to another retail planning role with the same optional field, the correct output is no letter and an hour spent moving forecasting evidence to the top of the resume. What would change this verdict: if the resume's most recent title already contained the word marketplace, the visible question disappears and situation five returns.

## Failure modes

**Resume Restated.** The letter walks through the same roles in the same order in prose. From the outside it reads as a document that took effort and conveyed nothing, and the reader stops at paragraph two because they have already read this.

**Mission Mirroring.** The company's own words about itself, handed back. It signals that one page was opened, and it is the single most common substitute for a real hook.

**Apology Spiral.** A gap or a switch explained across three paragraphs of context and mitigation. From the outside, the length is the confession: the reader concludes the candidate thinks this is disqualifying, and takes their word for it.

**Passion Substitution.** Enthusiasm inserted at exactly the points where evidence is missing. It is detectable by counting: the adjectives cluster around the requirements the resume cannot support.

**Wrong Company.** A reused letter that names the previous employer, usually in the second paragraph or the close, because the opening was edited and the rest was not. This is not rare. It is the standard failure of volume applying, and it ends that application immediately.

**Volume Decay.** One letter lightly edited for twelve applications. The hook goes generic first, then the evidence stops matching the posting, and by the eighth the letter argues for a role nobody advertised. Visible from the outside because the specifics point at nothing checkable.

**Overlength.** A 600-word letter for a required text box. It signals a misread of the situation before a word of content is assessed, and in the queue-worker situations it is scrolled past rather than read.

**The Unanswerable Objection.** Raising a concern the reader had not formed. A three-month gap between two jobs is invisible until you explain it, and then it is the thing the letter is about. Only answer questions the resume actually poses.

## What you hand over

Two things, in this order.

1. **One line:** the situation number, the reader it was written for, and the length band that follows. For example: "Situation 4, hiring manager who will see a domain switch on the resume, 200 to 300 words." Where the verdict is no letter, this line is the entire deliverable and it says why.
2. **The letter**, with the addressee resolved, the slots filled to budget, and a word count at the end.

## What this skill does not do

- **It cannot tell you whether a specific employer reads cover letters.** Nothing published breaks reading rates down by company or by team. The classification is inference from how the letter arrives.
- **It cannot make a weak application strong.** If the resume does not evidence the requirements the posting treats as non-negotiable, no letter closes that distance, and three unevidenced items argued well is still three unevidenced items.
- **It will not write an emotional appeal.** No hardship narrative, no urgency, no appeal to the reader's sympathy. Some hiring situations reward one and this file is not the place to get it.
- **It does not handle the follow-up.** Timing, chasing, the thank-you note and the reply to a rejection are all outside it.
- **It is not a substitute for a referral.** Where you have one, the letter matters much less, and the correct use of this file in that case is situation three, which is the shortest thing it produces.
- **It does not touch the resume.** Content, layout, parsing and file format are somebody else's job, and a good letter attached to a document that failed to parse changes nothing.
Why import instead of copy

A skill is only as good as what it can read.

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.

Start free