hiring-brief

---
name: hiring-brief
description: Turns "we need someone to do X" into a job advert built on 4 to 6 measurable performance objectives, plus a weighted scorecard for sifting applicants. Use for any job advert, role brief or hiring scorecard.
---

# A job advert that attracts the person you actually want

The advert, plus the scoring sheet that says who to interview.

## Claude does

1. Ask for five inputs and guess none: the job title; what goes wrong today because nobody does this job; the salary range; location, hours and working pattern; the hiring manager it goes out under. No range, stop and ask. An invented one is a false promise.
2. Write the mission in one sentence: why the role exists, stated as the result, not the tasks (ghSMART scorecard, part one).
3. Write 4 to 6 Key Performance Objectives, no more and no fewer (Adler). Each is Task + Action + Result, with a measurable result and a timeframe. Sequence them chronologically: first 30 days understand the problem, next 30 days build the plan, then execute.
4. Delete the filler. Adler's method replaces the skills list, the years-of-experience line, the behavioural adjectives and the degree requirement. Anything not traceable to a KPO is cut.
5. Write 4 to 6 competencies, each with the observable behaviour that proves it.
6. Build the scoring frame: 5 to 10 criteria, weights summing to 100%, anchored 1 to 5 (5 significantly exceeds, 4 exceeds, 3 meets, 2 partially meets, 1 fails). Mark 1 or 2 knockouts, where scoring below 2 eliminates the candidate whatever the total.

## Then Codex does

Codex does the repetition: the role as a job-board advert, a LinkedIn post and a screening set, plus the scorecard CSV and a banned-phrase lint. Same facts across channels, not judgement.

```
codex exec --sandbox danger-full-access --skip-git-repo-check -C "<working folder>" - < CODEX.md
```

Claude fills these first: {ROLE_BRIEF} (mission, KPOs, competencies), {COMPANY_FACTS} (company, location, hours, salary range, closing date, hiring manager), {SCORECARD} (criteria, weights, knockouts).

## Claude checks

1. Count the KPOs. Under 4 or over 6, reject and rerun.
2. Every KPO carries task, action and a measurable result with a timeframe. A missing result is the usual failure. Reject that row.
3. Search every draft for the phrases banned in CODEX.md. Any hit, reject.
4. Weights total exactly 100%, every knockout is in the CSV, and the advert names its hiring manager.
5. Every number traces to something the user supplied. A salary or growth figure from nowhere is invented. Reject and rerun.

## Rules

- 4 to 6 KPOs. Not 3, not 12.
- Cut anything a candidate cannot be scored on.
- Never present the "90% hiring success" figure as independent research. It is the publisher's own claim.
- Knockout beats the total: a 1 on a must-have is out, however high the sum.
- Public information only.
- Never invent a fact, a number or a quote.
- Anything sent in someone's name says whose name it is.

## If Codex is not installed

Claude writes the adverts, questions and CSV itself, and says so.

## Built from

- Lou Adler on performance-based job descriptions, https://www.preemploymentassessments.com/blog/performance-based-job-descriptions-and-kpos-with-lou-adler/ - the 4 to 6 count, Task + Action + Result.
- ghSMART, Who: The A Method for Hiring, https://ghsmart.com/insights/who-the-a-method-for-hiring/ - the mission/outcomes/competencies scorecard.
- AuraVMS, Weighted Quote Scoring Matrix, https://www.auravms.com/blogs/weighted-quote-scoring-matrix-supplier-evaluation - the 1 to 5 anchors, knockout rule.
- w95/awesome-claude-corporate-skills, https://github.com/w95/awesome-claude-corporate-skills - 189 stars via api.github.com.
- Anthropic, Skill authoring best practices, https://platform.claude.com/docs/en/agents-and-tools/agent-skills/best-practices - the format spec.

# A job advert that attracts the person you actually want - Codex task

## You are given

- {ROLE_BRIEF} - the approved brief: a one-sentence mission, 4 to 6 Key Performance Objectives each written as Task + Action + Result with a timeframe, and 4 to 6 competencies with their observable behaviours. Do not rewrite the objectives.
- {COMPANY_FACTS} - company name, location, hours and working pattern, salary range, closing date, how to apply, and the hiring manager's name the advert is sent under.
- {SCORECARD} - 5 to 10 criteria, their weights summing to 100%, and which of them are knockouts.

## Produce

Four files in the working folder.

1. `advert-jobboard.md`, 350 to 500 words. Order: role title; mission; "What you will have done by day 90" listing the KPOs verbatim; competencies; salary range; location and hours; how to apply; closing date; signed with the hiring manager's name.
2. `advert-linkedin.md`, under 200 words. Same facts, written first person from the hiring manager, no more than 3 hashtags.
3. `screening-questions.md`, exactly 6 questions, numbered, each labelled with the KPO number it tests. One per KPO where there are 6; where there are fewer, ask twice on the earliest KPOs and say which.
4. `scorecard.csv`, header `criterion,weight_pct,knockout,anchor_5,anchor_4,anchor_3,anchor_2,anchor_1`. One row per criterion. Anchors are 5 significantly exceeds, 4 exceeds, 3 meets, 2 partially meets, 1 fails, each written for that specific criterion rather than copied. `weight_pct` must total 100. `knockout` is yes or no.

## Rules

- Every fact comes from {COMPANY_FACTS}. If a fact is missing, write `MISSING: <field>` in the file. Never fill a gap with a plausible value.
- Banned from every draft: "years of experience", "degree required", "rockstar", "ninja", "guru", "hit the ground running", "wear many hats", "work hard play hard".
- No gendered or age-coded wording, including "young", "energetic" and "digital native".
- Add no requirement that is not in {ROLE_BRIEF}.
- Keep the KPO wording exactly as given in the job-board advert. Reword only for LinkedIn, and keep the measurable result and the timeframe in each.
- Invent no salary, headcount, growth figure, award or quote.
- British English throughout. No em dash characters.

## Return

Print, in this order:

1. The file names written.
2. The KPO count found in {ROLE_BRIEF}.
3. The `weight_pct` total.
4. The list of knockout criteria.
5. The result of searching every draft for each banned phrase, one line each, as `phrase: 0` or the offending line.
6. Every `MISSING:` line written, or `MISSING: none`.
7. The word count of each advert.