/job-description-generator

Purpose

/job-description-generator

What it IS: a structured first pass that produces a JD draft a human recruiter can review, edit, and own, with every bias-adjacent choice made explicit as a finding rather than buried in the prose.

What it is NOT: an Automated Employment Decision Tool (AEDT) under NYC Local Law 144, a candidate-facing system, a resume screener, an interview guide, a compensation analysis, or a substitute for employment counsel on jurisdiction-specific pay-transparency or protected-class questions.

This skill is governed by

hr-ai-governance

When to Use

Invoke

/job-description-generator

• Open a brand-new role and need a JD draft that starts from neutral language rather than a lifted template
• Refresh an existing JD that was authored before pay-transparency rules landed in the deployment jurisdiction
• Check an externally-drafted JD for gender-coded language, age-coded language, or required-vs-preferred bloat before posting
• Convert a legacy JD from a jurisdiction without pay transparency to one that requires it (e.g., posting a role in NYC that was originally written for Texas)
• Validate that a JD's stated seniority ("Senior," "Lead," "Staff") is consistent with the comp band the hiring manager expects

When NOT to Use

Do NOT use

/job-description-generator

• You need to screen, summarize, rank, or score incoming resumes against the JD → use

  /resume-summarizer

  (explicit AEDT wall; see

  hr-ai-governance

  Section 6)
• You need to build the interview guide that evaluates candidates against this JD → use

  /interview-guide

• You need compensation analysis, pay-gap audit, or salary-band construction → use

  /comp-benchmark

• You need performance evaluation criteria or a PIP framework → out of scope; see

  /performance-specialist

  consultations
• You need org design or headcount planning — out of scope; see

  @chro

• You need a legal opinion on whether a specific JD requirement is enforceable under a specific state's employment law → out of scope; engage

  @employment-counsel

The skill deliberately sits UPSTREAM of

/interview-guide

/comp-benchmark

Modes

Create (default)

Generate a fresh JD from scratch given role, jurisdiction, essential functions, seniority level, and (if pay transparency applies) comp band.

/job-description-generator --role "Customer Success Lead" --jurisdiction US-DE
/job-description-generator --role "Senior Backend Engineer" --jurisdiction US-NYC --comp-band comp-bands/backend-senior.json

Update

Refresh an existing JD. Pass the path to the current JD file.

/job-description-generator update Legionis/Product/jd-customer-success-lead.md --jurisdiction US-NYC

Update mode re-runs all scanners against the current JD, preserves section structure where possible, and produces a diff-aware findings list (new findings marked

[NEW]

~~strikethrough~~

Check

Scan an externally-drafted JD (one a hiring manager wrote on their own) without regenerating it. Produces findings only, no rewrite.

/job-description-generator check path/to/manager-draft.md --jurisdiction US-CA

Required Inputs

The skill MUST collect the following before producing output. If any required input is missing, it asks the user rather than guessing.

US-DE

US-NYC

US-CA

US-CO

US-WA

US-IL

EU-DE

UK

IL

$130,000–$170,000 base + equity

Legionis/Product/jd-customer-success-lead.md

Jurisdiction is never defaulted. If the user does not supply it, the skill asks. The jurisdiction determines pay-transparency obligations, the applicable framework under

hr-ai-governance

Essential functions are never inferred. The skill does not guess what a "Customer Success Lead" does; it requires the user to supply the essential functions, because essential functions are the ADA-relevant foundation of the JD. Inferred essential functions are a disability-discrimination exposure vector.

Output Structure

Every

/job-description-generator

sensitive-skill-guardrails.md

1. Disclaimer + UPL Guardrail Block (top)

> ⚠️ **Not HR or employment-law advice.** This output is a drafting and triage aid generated by a product-organization skill, not HR counsel or employment counsel. No attorney-client relationship is created by its production or use. Jurisdiction-specific questions on pay transparency, protected-class signaling, essential-function framing, ADA compliance, or enforceability require review by a licensed employment attorney in the relevant jurisdiction and by a qualified HR professional. Do not rely on this output as the sole basis for posting, recruiting, interviewing, or any employment decision.
>
> **Jurisdiction Assumed:** {jurisdiction from the required input}. If your jurisdiction differs, treat every finding below as a hypothesis to verify with local counsel.

2. JD Metadata block

A small labeled block under the disclaimer with:

• Role: title as input
• Jurisdiction: as input (no default)
• Seniority Level: IC2 / IC3 / IC4 / IC5 / Manager / Senior Manager / Director
• Comp Band: min/mid/max as input, or

  [not supplied — jurisdiction does not require pay transparency]

• Pay Transparency Required:

  yes

  /

  no

  (derived from Section 8.1 jurisdiction matrix + Section 2 of this skill)
• Pack Inheritance:

  hr-ai-governance v{version}

  (read at runtime)
• Generation Timestamp: ISO 8601 UTC
• Governance Pack Version:

  hr-ai-governance v1.0.1+

• Scanner Wordlist Version: Gaucher et al. 2011 + non-obvious framing extensions (see Method Step 5)

3. The JD Draft

Structured sections in this order:

1. About the role — 2-3 sentence framing, neutral language, no qualitative modifiers added by the model on its own authority
2. Essential functions — numbered list, one function per item, each item expressed as an observable activity ("Lead weekly customer health reviews" not "Own customer success")
3. Preferred qualifications — distinct from essential functions, explicitly labeled as preferred, each with an "or equivalent experience" alternative where a degree is listed
4. How we work — work arrangement, timezone expectations, collaboration style (no culture-fit language; culture ADD framing)
5. Compensation — pay range if jurisdiction requires it; explicit statement of equity / benefits if applicable; omitted if jurisdiction does not require and user did not supply a band
6. Equal opportunity statement — neutral, jurisdiction-appropriate language (US → EEOC; EU → GDPR + AI Act notice; IL → PPL-appropriate)

Every line of the JD is the skill's structured extraction of the user's input. The skill does NOT add adjectives, qualitative modifiers, or unsupported claims. If the user's input does not support a line, the skill does not write the line.

4.

## Scanner Findings

Every flagged term gets a row in a table with:

• Finding #
• Category — gender-coded (masculine) / gender-coded (feminine) / age-coded / disability-coded / citizenship-coded / culture-fit / degree-no-alternative / comp-band-title-mismatch / pay-transparency-missing / essential-preferred-bleed / qualitative-modifier-inserted / structural-finding
• Term or passage
• Where it appears — section of the JD
• Severity — P0 / P1 / P2
• Why it matters — the risk or implication
• Suggested neutral alternative — or "remove" for terms without a neutral replacement
• Verdict — address / accept-with-risk / reject-as-hypothetical

The scanner findings are the core bias-reduction output of the skill. They are how the hiring manager sees what the skill flagged and what it suggested. Structural findings (empty essential functions list, comp band not matching stated seniority, missing pay range in a pay-transparency jurisdiction) appear in the same table with

structural-finding

5.

## Essential vs. Preferred Classification Report

A two-column table showing what the skill classified as essential vs. preferred, with a one-line rationale for each. The hiring manager reviews this table before approving the JD, because the 8/8 rule (over-requiring narrows the pipeline disproportionately for groups that apply only when they meet 100% of requirements) makes this classification load-bearing. Items that appear in "essential" get a follow-up question: "Is this truly essential, or can a candidate learn it in the first 90 days?"

6.

## Comp Band Alignment Report

Co-owner contribution from 💵 Compensation Analyst. If a comp band was supplied, this section records:

• The stated seniority level
• The supplied comp band (min / mid / max)
• The alignment verdict:

  aligned

  /

  title-band-mismatch

  /

  cannot-assess

• If

  title-band-mismatch

  : the specific conflict (e.g., "Title says 'Lead' which implies IC5+ but band midpoint is at IC3 market") and suggested resolution (adjust title OR adjust band)
• If

  cannot-assess

  : the reason (e.g., "No market data supplied for 'Customer Success Lead' at this seniority level in this geography — run /comp-benchmark first")

7.

## Pay Transparency Compliance Status

• Jurisdiction → pay transparency rule (e.g.,

  US-NYC → NYC DCWP pay-transparency required

  ,

  US-DE → no pay-transparency rule in force

  )
• Status:

  compliant

  /

  non-compliant

  /

  not-applicable

• If

  non-compliant

  : the skill REFUSES to emit a final JD. The output is a finding, not a draft. The user must supply a comp band before the skill proceeds.

8.

## Reviewer Checklist

Seven mandatory sign-off items before the JD is acted upon:

• Jurisdiction confirmed against role location and candidate residency
• Essential functions list reviewed by the hiring manager and confirmed ADA-defensible
• Every P0 scanner finding addressed or explicitly accepted-with-risk by named approver
• Essential-vs-preferred classification reviewed; any "essential" item that could be learned in 90 days reclassified as preferred
• Comp band alignment verdict reviewed (if supplied)
• Pay transparency compliance status verified against the jurisdiction
• Employment counsel engaged for items in "Cannot Assess Without" that apply to this role

9.

## Cannot Assess Without Licensed Counsel or Specialist

Minimum 5 items. Examples:

• Enforceability of non-compete clauses in {jurisdiction} — out of scope; requires employment counsel
• Actual market compensation data for {role} in {geography} — requires

  /comp-benchmark

• Team composition and culture fit — out of scope; culture fit is NOT a valid preference
• Individual hiring manager preferences that are not observable job-performance criteria — out of scope; not a valid JD input
• Competitive intelligence on other active JDs in the market — out of scope
• Visa sponsorship policy and immigration law — requires employment counsel + immigration counsel
• ADA reasonable-accommodation boundaries for the essential functions — requires employment counsel for any specific candidate accommodation request

10. Audit Log Entry

Emitted per

hr-ai-governance

• run_id

  ,

  timestamp

  ,

  skill_name

  ,

  skill_version

  ,

  governance_pack_version

• jurisdiction

• inputs_hash

  — SHA-256 of the canonicalized role-input object (role, jurisdiction, essential-functions, seniority, comp-band-hash). NEVER raw candidate data — this skill doesn't see candidate data.

• input_type: job_description

• proxies_detected

  — from the scanner findings, each flagged term becomes a proxies_detected entry

• redactions_applied

  — N/A for this skill (this is a generation skill, not an extraction skill)

• hitl_reviewer

  — authenticated identity of the recruiter or hiring manager who will review

• hitl_decision

  ,

  hitl_timestamp

  ,

  hitl_rationale

  — populated at HITL gate

• downstream_action

  — e.g.,

  published-to-careers-page

  ,

  held-for-revision

• signoff_path: standard

• retention_expiry

  — computed from jurisdiction per Section 4.3

The skill refuses to run if the deployer's auth system cannot supply an authenticated

hitl_reviewer

Method

The skill's generation flow is a 7-step pipeline. Order matters because later steps depend on earlier ones.

Step 1 — Load Jurisdiction Rules

Read the jurisdiction from the required input. Look up the applicable row in

hr-ai-governance

• Pay-transparency requirement (yes / no)
• Pay-transparency specifics (salary range required on posting, or range on request, or range in interview process)
• Protected-class signaling rules specific to the jurisdiction (e.g., CA and NY have broader protected classes than US federal baseline)
• Runtime warning block to emit (e.g., EU → Section 7.3 EU AI Act high-risk notice)
• Retention period for the audit log entry

If the jurisdiction is not in the matrix, the skill blocks and routes to @employment-counsel per Section 11.1.

Step 2 — Accept Role Requirements; Enforce Essential-vs-Preferred Separation

Read the essential functions list. For each item:

• Is it phrased as an observable activity? (verb + object, not an abstract noun)
• Is it something the skill can verify against other inputs, or is it abstract ("must be a team player")?
• Could it be learned in the first 90 days of the role? If yes → flag as

  essential-preferred-bleed

  , suggest reclassification as preferred

The skill does NOT add essential functions the user did not supply. Inferred essential functions are an ADA-compliance exposure vector.

Step 3 — Generate JD Draft with Neutral Source-Faithful Language

Write the JD sections in order. Every line is grounded in a specific user input. The skill does NOT add qualitative modifiers on its own authority (per

hr-ai-governance

If the user supplied a modifier explicitly ("we need someone with significant experience"), the skill retains it verbatim and flags it as a finding ("significant" is a qualitative modifier supplied by the user; consider specifying a minimum year count instead).

Step 4 — Run Gender-Coded Language Scanner

Wordlist source: Gaucher, Friesen, and Kay (2011), "Evidence That Gendered Wording in Job Advertisements Exists and Sustains Gender Inequality," Journal of Personality and Social Psychology. Public academic citation, not a vendor tool.

Masculine-coded (flagged as

gender-coded (masculine)

aggressive, ambitious, analytical, assertive, athletic, autonomous, boast, challenging, competitive, confident, courageous, decide, decisive, determined, dominant, driven, fearless, forceful, greedy, headstrong, hierarchical, hostile, impulsive, independent, individualistic, intellectual, lead, leader, logic, logical, objective, opinionated, outspoken, persistent, principled, reckless, self-confident, self-reliant, self-sufficient, stubborn, superior

Feminine-coded (flagged as

gender-coded (feminine)

affectionate, cheer, collaborate, committed, communal, compassionate, connect, considerate, cooperative, dependable, emotional, empathetic, feel, flatterable, gentle, honest, interpersonal, interdependent, kind, kinship, loyal, modest, nag, nurturing, pleasant, polite, quiet, responsible, sensitive, submissive, supportive, sympathetic, tender, together, trust, trustworthy, understanding, warm, whiny, yield

For each match, the scanner flags the term, records its location in the JD, suggests a neutral alternative (e.g., "leader" → "owns outcomes," "collaborative" → "works across functions," "aggressive" → "outcome-focused"), and sets severity based on context: P1 if the term appears in essential functions or required qualifications; P2 if it appears in the "about the role" framing.

Balance rule: if masculine-coded count exceeds feminine-coded count by more than 3, OR vice versa, the overall JD is flagged P0. The empirical finding (Gaucher et al.) is that lopsided language — particularly masculine-coded imbalance — measurably reduces application rates from the under-represented gender.

Step 5 — Run Non-Neutral Framing Scanner

Beyond the gender-coded wordlist, the skill scans for:

hr-ai-governance

Step 6 — Comp Band Alignment Check (co-owner contribution)

I consulted 💵 Compensation Analyst on the title-to-band consistency logic. The alignment check:

1. Read the stated seniority (e.g., "Senior," "Lead," "Staff," "Principal")
2. Read the supplied comp band (min / mid / max)
3. Map the seniority to an expected band range (this mapping is owned by @compensation-analyst and lives in the comp-benchmark pack; this skill consumes it as a dependency)
4. If the supplied band is > 20% below the expected range for the stated seniority → flag

   title-band-mismatch

   P0: "Title says {X} which implies IC{N}+ market rate; supplied band midpoint is at IC{N-1}. Either adjust the title to match the band, or adjust the band to match the title."
5. If the supplied band is > 20% above the expected range → flag

   title-band-overpay

   P1: "Band is above market for stated seniority. Confirm this is deliberate (retention hire, scarce skill, high-cost geography) or adjust."
6. If market data is not available for the specific role + geography → return

   cannot-assess

   and surface "Run /comp-benchmark first" as the suggested next step.

The alignment check is load-bearing because title inflation (calling a mid-level role "Senior" or "Lead" without the comp to match) is a well-known hiring anti-pattern that wastes recruiter cycles on candidates who will reject the offer once the band is disclosed — which, under pay-transparency laws, happens at application time, not at offer stage.

Step 7 — Pay Transparency Compliance Check

Per the jurisdiction loaded in Step 1:

• If the jurisdiction requires pay-range-on-posting: verify a comp band was supplied; verify it appears in the Compensation section of the JD draft; verify the format matches the jurisdiction's rule (e.g., CA requires "good faith" range; NYC requires min and max).
• If the jurisdiction requires pay-range-on-request: the skill notes the obligation and advises the recruiter to have the range ready for any candidate who asks.
• If the jurisdiction does not require pay transparency: the skill does NOT require a comp band as input, BUT if a comp band is supplied, it still runs Step 6 alignment.
• If compliance status is

  non-compliant

  : the skill REFUSES to emit a final JD. The output becomes a finding: "Cannot generate final JD for {jurisdiction} without comp band."

Quality Gates

The skill performs a 10-item self-check BEFORE emitting output. If any check fails, the output does not publish — it produces a structural finding instead, asking the user to fix the gap.

1. Jurisdiction declared in the metadata block
2. Essential functions list is non-empty and distinct from preferred
3. No masculine-coded / feminine-coded imbalance > 3 terms
4. No age-coded, disability-coded, or citizenship-coded red flags at P0
5. No culture-fit language — culture-add framing only
6. Every degree requirement has an "or equivalent experience" alternative
7. Pay range present if the jurisdiction requires pay transparency
8. Comp band (if supplied) matches the stated seniority level within the tolerance band
9. No qualitative modifiers inserted by the model on its own authority (

   hr-ai-governance

   Section 6.2 #8 enforcement)
10. Audit log entry generated with a valid

    hitl_reviewer

    identity

These 10 checks are the quality gates. They run every invocation. There is no "express mode" that skips them.

Pack Inheritance

This skill inherits the following from

hr-ai-governance

inputs_hash

jurisdiction

hitl_reviewer

signoff_path

retention_expiry

qualitative-modifier-inserted

/resume-summarizer

This skill is NOT an AEDT. It produces a JD draft, not a candidate decision. The AEDT wall from

hr-ai-governance

Delegation Patterns Available

Default: Pattern 1 Consultation

When a specific element of the JD needs specialist input, the skill spawns a consultation per

delegation-protocol.md

Consultations are attributed in the scanner findings section or the alignment report: "I consulted 💵 Compensation Analyst, who noted that the supplied $130-170k band is ~15% below the market midpoint for a 'Lead' title in remote-Israel + US-Eastern in Q2 2026, and recommended either retitling to 'Senior' or increasing the band." Ownership of the JD draft stays with Recruiter.

Adversarial Review

NOT applicable at this version. Adversarial Review (Pattern 5) is reserved for near-final deliverables with high-stakes, uncapped exposure (enterprise contracts, M&A documents, pricing commitments). A JD draft is neither near-final nor uncapped-exposure at the generation stage; it is a draft going to a human reviewer who will revise and publish.

If a specific JD becomes high-stakes (e.g., a senior executive search with public press attention, or a role with ITAR/clearance requirements), the hiring manager can request

/job-description-generator --escalate-adversarial

@employment-counsel

ROI Framing

ROI for

/job-description-generator

HR blended rate: $150/hr per

feedback_roi_rates.md

Time-saved baseline: a careful, bias-scanned, jurisdiction-aware JD draft that integrates comp band alignment and produces a scanner findings table is ~2-3 hours of manual drafting and triage time for a recruiter working with a hiring manager. Simpler refreshes (Update mode on an existing JD with no pay-transparency change) are ~0.75-1 hour baseline. Complex roles (novel function, multi-jurisdiction posting, regulated industry with additional counsel consultation) are 4+ hours baseline.

The ROI tracks ONLY the time the skill saves on drafting the structured artifact, not the substantive HR review time. The substantive review by the recruiter, hiring manager, and (where needed) employment counsel still happens in full.

Example ROI line for a standard JD generation:

⏱️ ~2.5 hrs saved on drafting and triage in 45s, 15k tkns ~$0.9 cost, Value ~$375

Attribution and Maintenance

Owner: 🎯 Recruiter. The skill's drafting and scanner logic is Recruiter's accountability.

Co-owner: 💵 Compensation Analyst. Joint accountability on the comp band alignment logic in Method Step 6 and the Comp Band Alignment Report output section. Any change to the title-to-band tolerance thresholds, the alignment verdict taxonomy, or the

cannot-assess

Consumers (skills / gateways that invoke this skill):

• ext-hr

  — HR team gateway, primary user

• interview-guide

  — downstream skill; the JD defines the evaluation dimensions for the interview guide

• comp-benchmark

  — upstream when the user runs comp-benchmark first to get a market band, then feeds it into JD generation

New consumers require a frontmatter update and a one-line note in the consuming gateway's or skill's dependency list.

Authoring: First-principles. This skill was authored from scratch during Phase 4A as the first HR skill under the

hr-ai-governance

Dependency on the pack: The skill reads from

hr-ai-governance

Updates: Via the two-pass publication gate defined in

sensitive-skill-guardrails.md

/resume-summarizer

Minor edits (typos, formatting, scanner wordlist additions sourced from the pack's proxy register updates) can bypass Pass 2. Any edit touching: severity thresholds, scanner category taxonomy, comp band alignment logic, pay transparency compliance logic, or the HITL gate placement — requires a full Pass 2 substantive review by Employment Counsel.

Changelog: Maintained at the bottom of this file.

Example Invocation

User: /job-description-generator --role "Customer Success Lead" --jurisdiction US-DE --comp-band "$130-170k base + equity" --essential-functions legionis/cs-lead-essential.md

/job-description-generator v1.0.0 — loading:
  - Jurisdiction: US-DE (no pay-transparency rule; comp band supplied voluntarily)
  - Governance pack: hr-ai-governance v1.0.1
  - Scanner wordlist: Gaucher et al. 2011 + non-obvious framing extensions
  - Essential functions: 6 items (from cs-lead-essential.md)
  - Seniority: Lead (IC5-equivalent per user input)
  - Comp band: $130-170k base + equity

Running pipeline:
  Step 1: Jurisdiction rules loaded (US-DE: no PT requirement; 3-year retention default)
  Step 2: Essential functions parsed (6 items; 1 flagged as essential-preferred-bleed)
  Step 3: JD draft generated (neutral source-faithful language; 0 model-inserted modifiers)
  Step 4: Gender-coded scanner (3 masculine terms flagged, 2 feminine terms flagged — within balance)
  Step 5: Non-neutral framing scanner (1 culture-fit term flagged, 0 age-coded, 0 disability-coded)
  Step 6: Comp band alignment (consulted @compensation-analyst: Lead title + $130-170k = title-band-mismatch P0)
  Step 7: Pay transparency compliance (US-DE: not-applicable; comp band displayed as optional)

Producing output at:
  Legionis/Product/job-description-generator-birth-test-2026-04-11.md

10/10 quality gates passed (after title-band-mismatch finding surfaced for hiring manager decision).

⏱️ ~2.5 hrs saved on drafting and triage in 45s, 15k tkns ~$0.9 cost, Value ~$375

Changelog

• 1.0.0 (2026-04-11) — Initial authoring. First-principles during Phase 4A as the first HR skill under the

  hr-ai-governance

  pack. Co-owned by 🎯 Recruiter (primary) and 💵 Compensation Analyst (comp band alignment). Scanner wordlists from Gaucher et al. 2011 (public academic source) +

  hr-ai-governance

  Section 3.2 proxy register. Subsequent-similar 72-hour Pass 2 review under Employment Counsel SLA (the pack carries first-of-type status). Scaffolding review by 📋 Director of HR. Birth-tested against a Legionis Customer Success Lead role (see

  Legionis/Product/job-description-generator-birth-test-2026-04-11.md

  ).