12 Claude Skills for an Employer Brand the First Week Does Not Disprove

What these skills are

An employer brand with a living witness behind every claim. A firm of 20 to 200 people cannot spend its way to an employer brand, and does not need to. What it has instead is a credibility problem: the careers page describes the firm the founder wishes existed, the job ad is last year's with the title changed, and the joiner finds out in week two. The words were not written from anything.

One line holds through all of it: every claim about working here needs a living witness — a current colleague said it, or a founder signs it.

No skill in this pack invents that witness. The synthesizer reads the notes from your own listening round and nothing else about individuals. The EVP tags every claim with how strong the evidence behind it is, and what you cannot evidence yet goes on a list you fix in the workplace rather than in the copy. Nobody is quoted who has not approved the quote, and a page with no quotes at all is fine.

Mechanically, each skill is one folder with a SKILL.md file. Three gather the evidence, one writes the promise, five tell it where candidates read, and three test whether it held. They work from one Claude Project per firm with a pinned chat per unit of work, and each reads what the ones before it wrote, by file name — so what accumulates over a year is the firm's employer brand as a record rather than a poster.

Download all 12 skills. One zip: ready-to-install skill zips, readable SKILL.md files, and the three templates the pack runs on — the firm context, the interview notes, and the joiner's first-weeks check. Free, no signup. The same 12 skills are open on GitHub: polar-bear-org/claude-skills.

The 12 skills you get

The set follows one loop: three skills gather the evidence, one writes the promise, five tell it where candidates read, and three find out whether it held.

1 · Listen before you write a word

1. Evidence Interview Designer

Use when: You want to know what it is actually like to work here before you write a word about it

Output: The listening round designed: who to talk to (a spread across roles and tenure, not the fans), a stay-interview guide with open questions and follow-ups, the consent script, and the note-taking rule

2. Peer Claims Mapper

Use when: You have read ten agency careers pages and they all sound the same, and you want yours not to

Output: The claims on 5 to 10 peer firms' careers pages and job ads sorted into table stakes, rare and specific — a list of phrases your promise must not lean on, so it avoids the noise everyone makes

3. Evidence Synthesizer

Use when: The conversations are done and you have a pile of notes and no idea what they add up to

Output: The evidence base: themes with counts, the minority view kept, contradictions named, leadership view set against team view. Built from the filled interview notes and nothing else about individuals

2 · Write the promise you can keep

4. EVP From Evidence Writer

Use when: You need the one page that every job ad, careers page and post will be checked against

Output: The EVP as Give and Get pairs, each claim tagged with how strong the evidence behind it is, plus the list of what you will not claim yet — next quarter's to-do list for the workplace, not for the copy

3 · Tell it, where candidates actually read

5. Careers Page Writer

Use when: Your careers page is a stock photo and three adjectives, or you don't have one

Output: The page written from the EVP and nothing else: what it is like here, the hard parts said plainly, who should not apply, how hiring really works, pay policy, and only quotes people have approved

6. Job Ad Writer

Use when: A role just opened and you want twelve good applications, not two hundred

Output: One ad that attracts the right few: the real first 90 days, the give and the get for this role, "you'll dislike this if", the pay range, and the process with dates

7. Realistic Preview Builder

Use when: Candidates keep being surprised by the job in month two

Output: A realistic preview of one role, built from a conversation with the person doing the job today and approved by them: a written "a real week" piece and, if wanted, a short video script and shot list

8. Team Voice Editor

Use when: Someone on the team is willing to post about the work but doesn't know where to start

Output: Their post, still theirs: it interviews them, structures their own answers, marks every sentence it moved or bridged, and hands the draft back to them to finish

9. Hiring Content Planner

Use when: You post about hiring only when a role opens, and it shows

Output: A quarter of content where every piece maps to one EVP claim and one willing human owner, and shows evidence rather than asserting it. A piece with no owner is dropped, not assigned

4 · Find out whether the promise held

10. Brand Measure Designer

Use when: Someone asked "is any of this working?" and you had nothing to say

Output: The handful of measures that mean something when you hire a few people a year, and how to collect each with the least effort and no tracking of any individual

11. First Weeks Promise Checker

Use when: Someone starts next Monday and you want to know if what they were told is true

Output: Week 1, 4 and 12 conversations built from the exact claims the joiner read before they applied, and their approved answers turned into promise-kept and promise-broken flags on the hiring process. Nothing about the joiner

12. Brand Review Runner

Use when: A quarter has passed, people joined, one left, and the careers page hasn't moved

Output: The quarterly review: every EVP claim promoted or demoted on new evidence, every broken promise sorted into "fix the words" or "fix the workplace", and what changes next quarter, with an owner and a date

How the loop runs

One Employer Brand HQ project carries the whole thing. The filled firm-context.md sits in its files, and the loop opens in the Evidence chat: evidence-interview-designer sets up eight to twelve conversations, peer-claims-mapper runs alongside them on the pages your candidates read in the other tab, and evidence-synthesizer turns the filled interview notes into evidence-base.md.

evp-from-evidence-writer writes evp-give-and-get.md in the EVP chat, and that page is what everything after is checked against. careers-page-writer works from it and nothing else. job-ad-writer and realistic-preview-builder run in a "Role: [title]" chat, one per open role, so the ad, the preview and later that hire's first-weeks check all live together. hiring-content-planner plans a quarter in the Content chat, and team-voice-editor talks to the colleague directly — in their own chat if they would rather draft in private, which they should.

Then the promise gets tested. first-weeks-promise-checker opens a "Joiner: [first name]" chat the week before someone starts and comes back at week 1, 4 and 12 with the exact sentences that person read before they applied. brand-measure-designer sets the handful of measures once the page and the first ad are live. brand-review-runner closes the quarter: every claim promoted or demoted on new evidence, every broken promise sorted into fix the words or fix the workplace, with an owner and a date.

It stops where the job starts. The pack does not source, screen or interview candidates, negotiate offers, or benchmark pay — it insists you state a range and a real process, it does not set them. It does not design your onboarding program either: it tests the promise in a joiner's first weeks and hands the fixes back, which is where the Onboarding Pack picks up. And it writes the video script; it does not film it.

Setup guide

1. Download the pack. One zip: an install folder with 12 ready-to-upload skill zips, a skills folder with the same skills as readable files, and the three templates the pack runs on — firm-context.md, interview-notes.md and first-weeks-check.md.
2. Install your skills. In Claude Code, add the marketplace and install employer-branding-pack, and all 12 load at once. In Claude, turn on code execution in Settings, Capabilities, then go to Customize, Skills and upload one zip per skill from the install folder. Prefer working from files? Add the SKILL.md files to your Project knowledge instead; it works, just less cleanly.
3. Create your Employer Brand HQ. Make one Project per firm and call it Employer Brand HQ. Fill firm-context.md once — who the firm is, who it hires, pay policy, the hard parts it is willing to say out loud, and what it is not — and add it to the project's files. Every skill reads it, except team-voice-editor, which reads only the EVP's not-claimable list so a colleague's words stay their own.
4. Pin one chat per unit of work, and open Evidence. Name them exactly: Evidence, EVP, Careers page, Content, Measures, Review, plus one "Role: [title]" chat per open role and one "Joiner: [first name]" chat per new hire. Each skill saves under a fixed filename (evidence-base.md, evp-give-and-get.md, page-careers.md, ad-[role].md) and later skills read the earlier files by those names. Then open Evidence and write "run evidence-interview-designer". Everything else follows from the conversations it sets up.

Where to start

The quality bar

Every skill in the pack holds the same standard, the one we hold when we write about a workplace:

• Every claim about working here needs a living witness: a current colleague said it, or a named founder signs it
• No skill invents the witness, and nobody is quoted who has not approved the quote — a page with no quotes is fine, a page with a quote nobody said is the thing this pack exists to prevent
• A survey is not a listening round: eight to twelve real conversations, and the spread has to include the person who almost left last spring, not only the people who love it here
• The synthesis keeps the minority view and names the contradictions, because an EVP written from a flattering read is the promise the first week disproves
• What the firm cannot evidence yet goes on the not-claimable list: next quarter's to-do list for the workplace, not for the copy
• The best thing a job ad can do for the wrong candidate is help them not apply: the real first 90 days, who will dislike it, the pay range, the process with dates
• The team's posts stay the team's words — the editor interviews, arranges, marks what it moved, and hands the draft back; a post generated under someone's name is a press release wearing a face
• Content is planned around consent: a piece with no willing owner is dropped from the plan, never assigned
• The joiner's check produces flags on the hiring process, never on the joiner, and the joiner decides which of their answers the firm keeps
• A broken promise is sorted into fix the words or fix the workplace, and it is never closed by finding a nicer way to phrase it

Who made this

Polar Bear is a people ops consultancy for human-size teams (20 to 200 people). Built by ex-McKinsey founders with a dream to make AI work for People, not instead of them. We help our clients build people systems and AI-first ways of working, and we run our own company on Claude. This pack is the free, self-serve version of how we work.

The pack builds one firm's employer brand out of what its own people already say. When you want your whole company working this way, AI carrying the overhead so people do the thinking, across hiring, onboarding, and everyday work, that's what we build with clients.

Meet Pauline. Careers page you would rather not show a candidate? Bring it, the last job ad you posted, or the joiner who left in month three. Book a 30-minute call · Pauline on LinkedIn