Purpose
Job descriptions decay into a list of every task anyone ever did in the role, with
a degree requirement nobody checked and eight years of experience nobody costed.
This skill fixes the two things that actually change outcomes: which wording
suppresses applications from qualified people, and which requirements are load
bearing for the grade. It grades against explicit factors and refuses to output a
grade the supplied evidence does not support.
Data classification
Internal. A draft JD is internal until posted; the posted version is Public.
STOP CONDITIONS — check before doing anything else:
| Trigger |
Action |
| Named incumbent's performance data, salary, disciplinary history, or any employee PII is pasted |
Stop. State that JD authoring does not need it. Ask for it to be removed and for the role profile only. Do not restate the PII back. |
| The request is to write the JD around a specific named person so that only they qualify |
Stop. Say plainly that this is a recruitment-fairness risk and route to HR/legal. |
| Requirements are supplied that filter on age, sex, race, religion, disability, nationality, or family status |
Stop. Name the requirement. Do not draft it, do not soften it, do not "phrase it carefully". |
NON-FABRICATION RULE. Do not invent duties, reporting lines, budget size,
headcount, or a salary range. Do not infer the grade from the job title — titles
are the least reliable signal in the whole input set. If a grading factor has no
supporting evidence in the supplied material, output UNEVIDENCED — cannot grade
for that factor and stop before assigning an overall grade. A guessed grade
becomes a pay decision.
Prerequisites
- Inputs (minimum viable): the role's purpose in one sentence; 4–8 actual
accountabilities; who it reports to; what it is responsible for deciding.
- Inputs (needed for grading): the organisation's grading/job-evaluation
framework, or agreement to use the generic factor table below; two comparator
roles at known grades.
- Inputs (needed for compliance): which jurisdiction(s) the role will be
advertised in. Ask. Pay-transparency, salary-history, and accessibility-statement
duties differ by jurisdiction and this skill must not assume one.
- Access: none. No system access required.
If any minimum-viable input is missing, name it and stop. Do not draft around it.
Procedure
1. Establish jurisdiction and framework
Ask two questions before drafting: where will this be advertised and which
grading framework applies. If the user does not know the framework, use the
generic factor table in step 4 and label the output "indicative grade, not a
job-evaluation outcome".
2. Separate essential from desirable — this is the whole exercise
For every requirement, ask: could someone do this job on day 90 without it?
If yes, it is desirable. Requirements move to desirable by default; promotion to
essential needs a reason recorded in one clause.
| Requirement type |
Default |
Promote to essential only if |
| Degree |
Desirable |
Legally required, or a licensing body requires it |
| Named vendor certification |
Desirable |
Client contract or regulator mandates it |
| Years of experience |
Desirable |
Replace with a capability statement wherever possible |
| Specific tool/platform |
Desirable |
Nobody in the team can train it and the tool is unusual |
| Language fluency |
Desirable |
The role's daily work is conducted in that language |
| Licence, clearance, right to work |
Essential |
State it as a factual condition, not a preference |
Years of experience is the most common exclusion mechanism in a JD. "5+ years"
excludes career-changers and returners and correlates with nothing. Prefer
"can independently run a month-end close" over "3–5 years in a finance team".
3. Run the inclusive-language check
Flag every hit with the line, the reason, and a replacement. Do not silently
rewrite — show the user what was flagged.
| Category |
Flag |
Why |
Replacement pattern |
| Gendered nouns |
chairman, salesman, manpower, he/she as default |
Signals a default incumbent |
chair, salesperson, workforce, they |
| Aggressively masculine-coded |
dominant, competitive, aggressive, ninja, rockstar, guru, crush it, hard-charging |
Reduces application rates from women in tested wordlists |
leads, collaborates, delivers, skilled |
| Feminine-coded overload |
supportive, nurturing, compassionate used for a delivery role |
Narrows the field the other way |
name the actual behaviour |
| Age-coded |
digital native, recent graduate, young and energetic, high-energy team |
Proxy for age |
name the skill, not the life stage |
| Ableist / unexamined physical demands |
must be able to stand for long periods, must drive, strong verbal communicator |
Screens out disabled applicants when not essential |
state only genuine occupational requirements, and state them as tasks |
| Culture-fit language |
culture fit, like-minded, work hard play hard, family |
Licenses similarity bias |
culture add; name the values being tested |
| Availability-coded |
always on, thrives under pressure, flexible hours essential |
Screens out carers and disabled applicants |
state actual hours and any genuine on-call rota |
| Native-speaker requirements |
native English speaker |
Nationality proxy; usually unlawful |
"business-level fluency in X" |
| Unpaid or unbounded extras |
occasional weekends expected |
Hidden term |
state the contractual position |
Also check structurally:
- Is there a stated accessibility/adjustments line for the application process?
- Is the requirements list under 10 items? Long lists depress applications from
candidates who self-screen at partial matches.
- Is a salary or range present? Ask the user whether their jurisdiction requires
one; do not assume, and do not invent a figure.
4. Grade the role against factors
Score each factor only from evidence in the supplied accountabilities.
| Factor |
What the evidence must show |
If absent |
| Knowledge and skill |
Depth of expertise required, and how it is acquired |
UNEVIDENCED |
| Problem-solving |
Are problems precedented, adapted, or novel? |
UNEVIDENCED |
| Autonomy |
What is decided alone; what needs sign-off; what the escalation trigger is |
UNEVIDENCED |
| Scope of impact |
Team, function, organisation, or external |
UNEVIDENCED |
| Financial responsibility |
Budget owned or influenced, with a figure |
UNEVIDENCED — do not estimate |
| People responsibility |
Direct reports, matrix reports, or none |
UNEVIDENCED |
| Stakeholder complexity |
Internal only, or regulated/external counterparties |
UNEVIDENCED |
Stop rule: two or more UNEVIDENCED factors → no overall grade. Output the
factor table, the specific questions that would close each gap, and stop.
5. Assemble and self-check
Output in this order: purpose (1 sentence) · accountabilities (verb-led, 4–8) ·
essential requirements · desirable requirements · working arrangements ·
adjustments statement · grade evidence table.
Final check before handing back:
- Every essential requirement traces to an accountability.
- No requirement in the advert is absent from the JD, or vice versa.
- Flagged-language list is shown to the user, not quietly applied.
- Grade is either fully evidenced or withheld.
Boundaries
- Not for analysing development needs across an existing team — that is
hr-training-needs-analysis.
- Not for writing objectives or review wording for a current post-holder —
that is
hr-performance-review-coaching.
- Not for determining a salary range from market data; this skill grades the
role, it does not price it, and it will not invent benchmark figures.
- Not for immigration, sponsorship, or right-to-work determinations. Route to
qualified advisers in the relevant jurisdiction.
Hand-offs
- Receives from: a hiring manager's raw notes, or an existing JD needing refresh.
- Routes to:
hr-training-needs-analysis when the grading exercise reveals the
team lacks a capability the role assumes exists.
- Routes to:
hr-performance-review-coaching once a hire is in post and the
accountabilities need converting into SMART objectives.
- Routes to:
legal-compliance-policy-drafting if the JD exposes a missing
internal standard (e.g. no recruitment or adjustments policy exists).
1---2name: hr-job-description-authoring3description: Drafts and audits job descriptions against an inclusive-language checklist, an essential-versus-desirable split, and a job-evaluation grading rubric, then reports which grading factors are unevidenced rather than guessing a grade. Use when writing a new JD, refreshing an old one, checking a JD before it is posted, resolving a disagreement about which grade a role sits at, or auditing a set of adverts for exclusionary wording. Trigger on 'write a job description', 'review this JD', 'is this advert inclusive', 'what grade is this role', 'job spec', 'role profile', 'person specification'. Not for consolidating development gaps across an existing team (use hr-training-needs-analysis) or for writing objectives for someone already in post (use hr-performance-review-coaching).4---56## Purpose78Job descriptions decay into a list of every task anyone ever did in the role, with9a degree requirement nobody checked and eight years of experience nobody costed.10This skill fixes the two things that actually change outcomes: which wording11suppresses applications from qualified people, and which requirements are load12bearing for the grade. It grades against explicit factors and refuses to output a13grade the supplied evidence does not support.1415## Data classification1617**Internal.** A draft JD is internal until posted; the posted version is Public.1819**STOP CONDITIONS — check before doing anything else:**2021| Trigger | Action |22| --- | --- |23| Named incumbent's performance data, salary, disciplinary history, or any employee PII is pasted | Stop. State that JD authoring does not need it. Ask for it to be removed and for the role profile only. Do not restate the PII back. |24| The request is to write the JD *around* a specific named person so that only they qualify | Stop. Say plainly that this is a recruitment-fairness risk and route to HR/legal. |25| Requirements are supplied that filter on age, sex, race, religion, disability, nationality, or family status | Stop. Name the requirement. Do not draft it, do not soften it, do not "phrase it carefully". |2627**NON-FABRICATION RULE.** Do not invent duties, reporting lines, budget size,28headcount, or a salary range. Do not infer the grade from the job title — titles29are the least reliable signal in the whole input set. If a grading factor has no30supporting evidence in the supplied material, output `UNEVIDENCED — cannot grade`31for that factor and stop before assigning an overall grade. A guessed grade32becomes a pay decision.3334## Prerequisites3536- **Inputs (minimum viable):** the role's purpose in one sentence; 4–8 actual37 accountabilities; who it reports to; what it is responsible for deciding.38- **Inputs (needed for grading):** the organisation's grading/job-evaluation39 framework, or agreement to use the generic factor table below; two comparator40 roles at known grades.41- **Inputs (needed for compliance):** which jurisdiction(s) the role will be42 advertised in. Ask. Pay-transparency, salary-history, and accessibility-statement43 duties differ by jurisdiction and this skill must not assume one.44- **Access:** none. No system access required.4546If any minimum-viable input is missing, name it and stop. Do not draft around it.4748## Procedure4950### 1. Establish jurisdiction and framework5152Ask two questions before drafting: *where will this be advertised* and *which53grading framework applies*. If the user does not know the framework, use the54generic factor table in step 4 and label the output "indicative grade, not a55job-evaluation outcome".5657### 2. Separate essential from desirable — this is the whole exercise5859For every requirement, ask: **could someone do this job on day 90 without it?**60If yes, it is desirable. Requirements move to desirable by default; promotion to61essential needs a reason recorded in one clause.6263| Requirement type | Default | Promote to essential only if |64| --- | --- | --- |65| Degree | Desirable | Legally required, or a licensing body requires it |66| Named vendor certification | Desirable | Client contract or regulator mandates it |67| Years of experience | Desirable | Replace with a capability statement wherever possible |68| Specific tool/platform | Desirable | Nobody in the team can train it and the tool is unusual |69| Language fluency | Desirable | The role's daily work is conducted in that language |70| Licence, clearance, right to work | Essential | State it as a factual condition, not a preference |7172Years of experience is the most common exclusion mechanism in a JD. "5+ years"73excludes career-changers and returners and correlates with nothing. Prefer74"can independently run a month-end close" over "3–5 years in a finance team".7576### 3. Run the inclusive-language check7778Flag every hit with the line, the reason, and a replacement. Do not silently79rewrite — show the user what was flagged.8081| Category | Flag | Why | Replacement pattern |82| --- | --- | --- | --- |83| Gendered nouns | chairman, salesman, manpower, he/she as default | Signals a default incumbent | chair, salesperson, workforce, they |84| Aggressively masculine-coded | dominant, competitive, aggressive, ninja, rockstar, guru, crush it, hard-charging | Reduces application rates from women in tested wordlists | leads, collaborates, delivers, skilled |85| Feminine-coded overload | supportive, nurturing, compassionate used for a delivery role | Narrows the field the other way | name the actual behaviour |86| Age-coded | digital native, recent graduate, young and energetic, high-energy team | Proxy for age | name the skill, not the life stage |87| Ableist / unexamined physical demands | must be able to stand for long periods, must drive, strong verbal communicator | Screens out disabled applicants when not essential | state only genuine occupational requirements, and state them as tasks |88| Culture-fit language | culture fit, like-minded, work hard play hard, family | Licenses similarity bias | culture *add*; name the values being tested |89| Availability-coded | always on, thrives under pressure, flexible hours essential | Screens out carers and disabled applicants | state actual hours and any genuine on-call rota |90| Native-speaker requirements | native English speaker | Nationality proxy; usually unlawful | "business-level fluency in X" |91| Unpaid or unbounded extras | occasional weekends expected | Hidden term | state the contractual position |9293Also check structurally:94- Is there a stated accessibility/adjustments line for the application process?95- Is the requirements list under 10 items? Long lists depress applications from96 candidates who self-screen at partial matches.97- Is a salary or range present? Ask the user whether their jurisdiction requires98 one; do not assume, and do not invent a figure.99100### 4. Grade the role against factors101102Score each factor only from evidence in the supplied accountabilities.103104| Factor | What the evidence must show | If absent |105| --- | --- | --- |106| Knowledge and skill | Depth of expertise required, and how it is acquired | UNEVIDENCED |107| Problem-solving | Are problems precedented, adapted, or novel? | UNEVIDENCED |108| Autonomy | What is decided alone; what needs sign-off; what the escalation trigger is | UNEVIDENCED |109| Scope of impact | Team, function, organisation, or external | UNEVIDENCED |110| Financial responsibility | Budget owned or influenced, with a figure | UNEVIDENCED — do not estimate |111| People responsibility | Direct reports, matrix reports, or none | UNEVIDENCED |112| Stakeholder complexity | Internal only, or regulated/external counterparties | UNEVIDENCED |113114**Stop rule:** two or more UNEVIDENCED factors → no overall grade. Output the115factor table, the specific questions that would close each gap, and stop.116117### 5. Assemble and self-check118119Output in this order: purpose (1 sentence) · accountabilities (verb-led, 4–8) ·120essential requirements · desirable requirements · working arrangements ·121adjustments statement · grade evidence table.122123Final check before handing back:124- Every essential requirement traces to an accountability.125- No requirement in the advert is absent from the JD, or vice versa.126- Flagged-language list is shown to the user, not quietly applied.127- Grade is either fully evidenced or withheld.128129## Boundaries130131- **Not for** analysing development needs across an existing team — that is132 `hr-training-needs-analysis`.133- **Not for** writing objectives or review wording for a current post-holder —134 that is `hr-performance-review-coaching`.135- **Not for** determining a salary range from market data; this skill grades the136 role, it does not price it, and it will not invent benchmark figures.137- **Not for** immigration, sponsorship, or right-to-work determinations. Route to138 qualified advisers in the relevant jurisdiction.139140## Hand-offs141142- Receives from: a hiring manager's raw notes, or an existing JD needing refresh.143- Routes to: `hr-training-needs-analysis` when the grading exercise reveals the144 team lacks a capability the role assumes exists.145- Routes to: `hr-performance-review-coaching` once a hire is in post and the146 accountabilities need converting into SMART objectives.147- Routes to: `legal-compliance-policy-drafting` if the JD exposes a missing148 internal standard (e.g. no recruitment or adjustments policy exists).
alihusains