cover-letter
Skill2 starsChanged 3 months agoMITView on GitHub
Write a cover letter, recruiter reply, or application-form answer in the seeker's own voice — bare, two paragraphs, tailored to the posting itself and anchored to letters actually sent. Use whenever something is to be written in the first person as them.
36 other projects use a version of this.
Adopted by
Most recently taken up by
- carriegale2710/resume-tailortheir own version, first seen Nov 16, 2026
- Axect/skillstheir own version, first seen Nov 16, 2026
- Bellistri-Nick/pm-skillstheir own version, first seen Nov 15, 2026
- Muvon/octomind-taptheir own version, first seen Nov 12, 2026
- thesun4sky/jobstacktheir own version, first seen Nov 10, 2026
- eggs-gd/career-spacetheir own version, first seen Nov 8, 2026
Used with
Projects with it also use interview-prep, tailor-resume and 6 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
AnthropicSkillinterview-prep
Create structured interview plans with competency-based questions and scorecards. Trigger with "interview plan for", "interview questions for", "how should we interview", "scorecard for", or when the user is preparing to interview candidates.
Also in 8 projects that use this
proficientlyjobsSkilltailor-resume
Tailor your resume for a specific job posting
Also in 4 projects that use this
rhowardstoneSkilljob-search
Search for jobs across multiple boards and company career pages. Use when looking for new positions to apply to.
Also in 4 projects that use this
aslavchevSkillmock-interview
Simulate a real job interview for any role. Use when user says "mock interview", "interview me", "interview prep", "practice interview", or wants to rehearse for a specific company or position. Adapts rounds and questions to the role — technical, managerial, creative, or any other.
Also in 3 projects that use this
rhowardstoneSkillapply
Research-backed job application workflow emphasizing relationship-building over mass applications
Also in 3 projects that use this
JDerekLomasSkillliterature-review
Conduct comprehensive, systematic literature reviews using multiple academic databases (PubMed, arXiv, bioRxiv, Semantic Scholar, etc.). This skill should be used when conducting systematic literature reviews, meta-analyses, research synthesis, or comprehensive literature searches across biomedical, scientific, and technical domains. Creates professionally formatted markdown documents and PDFs with verified citations in multiple citation styles (APA, Nature, Vancouver, etc.).
Also in 3 projects that use this
Other versions36 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 36
- mergisi/awesome-openclaw-agents★ 4.1ksame name
- claude-office-skills/skills★ 433same name
- proficientlyjobs/proficiently-claude-skills★ 389same name
- thesun4sky/jobstack★ 216same name
- matthewdigiuseppe/MStack★ 17same name
- Mahashwetha/Job-search-email-system-claude★ 12same name
- pradyumnasagar/open-research-skills★ 8same name
- Muvon/octomind-tap★ 6same name
- xcrrr/claude-skills★ 6same name
- Axect/skills★ 4same name
26 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository11 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- apply-formSkill
- evaluate-roleSkill
- job-triageSkill
- publishSkill
- research-companySkill
- setupSkill
- sync-appliedSkill
- tailor-cv-batchSkill
- tailor-cvSkill
The file
We split this skill into the rules it gives an agent
9 rules. 0 are common, 7 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 32 edited versions we read keep these.
7
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 6
Write outbound prose as Ben.
- Distinctive: line 25
Read all the way to the end.
- Very specific: line 37
Run /research-company (engine: python -m research) unless it was already run for this role this cycle.
- Distinctive: line 45
Do not put the research in the letter.
- Very specific: line 83
If a tailored CV was built for this role, read its plan.json too — the letter must not contradict the résumé in the same envelope.
- Distinctive: line 92
No numbers: No user, subscriber or revenue counts — he rejected "500+ paying subscribers".
- Distinctive: line 111
Render the letter before showing it: Write the JSON, run the renderer, then put the body text in chat for Ben to reword.
- Distinctive: line 132
Do not assert anything about the posting you have not read: If a requirement is inferred from framing rather than quoted from the page, say …
- Distinctive: line 160
And this whole section is for a letter that was SENT: If Ben is exploring — asking for options, a few angles, something to paste into a …
“Common” is measured: kept by at least 6 in 10 of the 32 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.
Files
Write outbound prose as Ben. Not "professional writing about Ben" — his voice, which is far barer and shorter than the default. This covers recruiter replies, form free-text and outreach too, not just letters: the failure mode is identical in all of them.
Everything about voice and rules already exists in this repo. Read it — do not re-derive it, and do not add a parallel file for anything below.
0. Read the posting IN FULL. First. Before drafting a word.
This step is first because it was once skipped, and skipping it is the failure this skill exists to prevent. On 2026-07-27 a letter for Atlas Tech was drafted from the partial JD text the triage run had captured. The full posting said "US citizenship is a requirement for this position" and "ability to work onsite with customers approximately 25% of the time" — a hard gate and a logistics term, neither of them in the snippet, both cheap to answer in four words.
- Fetch the JD from the posting itself. A triage worklist entry, a
whyline, or an emailed digest is a snippet and is routinely truncated mid-requirement. If the page needs a login, ask for a paste. - Read all the way to the end. Requirements sections put the gating lines last.
- Pull out, explicitly:
- Gating requirements — citizenship, clearance, work authorization, location, onsite/travel percentage, employment type (1099 / W2 / perm). These are pass/fail and usually screened first.
- Core stack, in their words.
- The one thing they stress — the line that tells you what they actually care about. That is the tie-in candidate.
- Note which stack items have no backing in
profile/bullet-bank.md. Those are gaps to leave out of the letter and record in the application folder'sREADME.md— never claims to stretch.
1. Research the company
Run /research-company (engine: python -m research) unless it was already run for this role this
cycle. It answers things the posting will not: agency vs direct employer, what else is on their board,
whether Ben has a history with them, and whether the same req is being shopped by multiple agencies.
Two things this changes about the letter:
- A duplicate agency req means do not cold-apply — it can cut across a process already in motion.
- It supplies the honest, specific tie-in that keeps the letter from reading generic.
Do not put the research in the letter. It decides what to lead with, and that is all.
2. Tailor by SELECTION, not by accretion
This is the rule that resolves "keep it bare" against "tailor it to them." The letter stays two paragraphs whatever the target. Tailoring means choosing which true fact leads — not adding a sentence to cover another requirement.
- Their stack is Python-first? Lead the build sentence with Python.
- Mission-driven or defense? Lead paragraph 2 with the Air Force decade.
- AI-native product? Lead with training and running the models.
- Every requirement you "cover" by adding a clause makes the letter worse. If the letter is growing, you are answering the JD line by line, which is what makes a letter read as machine-written.
3. Answer the gating requirements — briefly, and only those
Gating requirements are the exception to bareness, because they are pass/fail and cost almost nothing: citizenship or work authorization, location, willingness to travel, employment type. One clause, folded into an existing sentence — never a new paragraph and never a bulleted list.
"I'm a US citizen, OKC based, but willing to travel to fulfill duties."
Everything else — a missing cloud, a missing database, a framework he has not used — stays out. Those
are conversation, and they belong in the application folder's README.md.
Sources of truth — read all of these, every run
| File | What it gives you |
| the posting itself | The requirements. See §0. Nothing below substitutes for it. |
docs/knowledge-base/personal/tailoring-playbook.md → "Cover letters — Ben's preferences" | The rules. The binding list. |
docs/knowledge-base/personal/ben-voice.md | The voice — verbatim samples and the traits distilled from them. |
profile/letters/ | The corpus — letters actually sent. Outranks the trait list wherever they disagree. |
profile/cover-letter.json | The standing letter. Copy it and edit; it is the shape a letter should be. |
profile/bullet-bank.md | The claims. DO-NOT-CLAIM applies to letters exactly as to résumés. |
If a tailored CV was built for this role, read its plan.json too — the letter must not contradict
the résumé in the same envelope.
The five rules that get broken most
In the playbook already; repeated here because each has been violated in a shipped draft.
- BARE, and no more than two paragraphs. Ben, verbatim: "anything more than this, it is obvious you are an AI." A longer letter is the failure mode, not the safe choice.
- No numbers. No user, subscriber or revenue counts — he rejected "500+ paying subscribers".
- The video goes inline in a sentence, never in a header or links block. Write it as a body segment list so "3-minute introduction video" is the clickable text. Add nothing to the header — Ben on a links line someone added: "this looks stupid."
- Never tell the employer about themselves. No "your team", no "the mission you're building for", no "your stack". State a fact about Ben and stop. Ben: "do not tell them how their team works — that is horrible." Note the tension with tailoring and resolve it the same way every time: tailoring changes which fact about him you lead with, never adds a claim about them.
- One JD tie-in, plainly — or none. A list of matches reads as AI. Own words, never mirror the JD.
- No interpretive payoff clause — state the fact and stop. A sentence that follows a concrete fact and explains what it meant is the loudest AI tell in the letter. Ben cut this one on 2026-07-27, after a paragraph naming Tinker, the Pentagon and the F100 engine: "That job was turning what the mission needed into what had to get built." His verdict: "this screams AI and i hate this sentence." Test: delete it. If the paragraph still says everything factual it said, it was interpretation — leave it out. The urge to write one comes from wanting to connect his experience to their posting; that connection is the reader's to make.
Build first, then show — do not gate on approval
Render the letter before showing it. Write the JSON, run the renderer, then put the body text in chat for Ben to reword. His rule, verbatim: "make it first just in case it is good and there is no delay." Most letters are close enough to send, and waiting for sign-off on a draft costs a round-trip every time.
Always paste the body text into the message — he rewords from chat, not by opening the docx — and
re-render on his edit. This is the same build-first default /tailor-cv uses.
After rendering — the AI-gloss pass, then the check
Run the AI-gloss pass before showing Ben anything — docs/knowledge-base/personal/tailoring-playbook.md →
"THE AI-GLOSS PASS". Open it; do not work from memory. A letter is more exposed to gloss than a CV
because it is prose, and the same nine tells apply. The three that recur here: negative parallelism
(not just X, it's Y), em-dash payoff clauses bolting a moral onto a fact, and defensiveness —
any sentence that reads as a reply to an objection nobody raised. Delete rather than rewrite; rewriting
is how a hedge becomes a different hedge. Say what was cut.
State plainly, in chat, in two lines:
- which gating requirements the posting names, and whether the letter answers each;
- which stack items they asked for that have no bullet-bank backing, so he goes in knowing the gaps.
Do not assert anything about the posting you have not read. If a requirement is inferred from framing rather than quoted from the page, say "inferred" — a confident guess about a hard requirement is worse than no guess.
Render it
cv/scripts/make_cover_letter.py is the renderer. There is no other one — do not write one.
The letter is JSON (date, salutation, body, closing); the name/title/contact header comes from
profile/profile.yaml → identity, not from the letter file.
.venv/bin/python cv/scripts/make_cover_letter.py \
--letter applications/<folder>/cover-letter.json \
--out applications/<folder>/<name>_cover_letter.docx --pdfA body paragraph is either a plain string or a list of segments, where a segment is a string or
{"text": ..., "url": ...} — that is how a link becomes clickable inline.
After it is sent
Append the letter to profile/letters/ as <YYYY-MM-DD>_<company-slug>.json. Reusable positioning
lessons go to docs/knowledge-base/personal/tailoring-playbook.md — write those unprompted.
A new verified fact about Ben goes to profile/bullet-bank.md, which is PROTECTED: read it freely,
ask before you write, every time. Propose it in chat with its evidence and apply it only on his yes.
A "yes" to write a letter is not a yes to edit the bank; see the notice at the top of that file.
And this whole section is for a letter that was SENT. If Ben is exploring — asking for options, a
few angles, something to paste into a form — the answer is the chat message and nothing else. Do not
create a file in the application folder for it. (2026-07-30, Sift: an agent turned "give me some
options" into a saved form-answers.md plus two documentation edits, none of them asked for.)
profile/ is classified PERSONAL by scripts/extract.py, so nothing there is ever published.
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
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.

