Prompts that run on governed role data
Copy any of these into Claude — or any agent connected to WorkforceGPT — and it will call the connector instead of guessing. Every prompt maps to something the server can actually do. The first five cost you nothing.
- Create a free WorkforceGPT account and verify your email.
- Add WorkforceGPT from Claude’s connectors directory, or give any other MCP client workforcegpt.ai/mcp.
- Approve the connection in your browser when prompted.
Most of what you can do here doesn’t use your allowance.
Nothing in this library is billed per use. Two actions draw on the allowance included with your account; everything else is unconstrained.
Listing your roles, opening a profile, searching published roles, revisiting past scores. No allowance used, no account needed for the public library.
Uses one assessment from your account’s allowance. The assessment output is detailed — get the most out of each one with prompts 06 and 07.
Uses one role generation. Generations run one at a time and finish in the background rather than instantly.
Allowances differ by account.
Rather than trust a number printed on a page, run prompt 01 — it asks the connector directly and reports what you have left right now.
Start without using anything
Five prompts that confirm the connection works and show you real output. None of them touch your allowance, and two of them work with no account at all.
01
Confirm the connection and check your allowance
NO ALLOWANCE USED
What comes back:
a plain-language inventory plus your live numbers. Run this first — if the agent answers from general knowledge instead of naming connector capabilities, the connection isn’t active.
02
See whether the role already exists
NO ALLOWANCE USED
What comes back: published role profiles, free to read, no account required. The fastest way to see the output before you commit to anything.
03
See what you already have
NO ALLOWANCE USED
What comes back: your own role inventory. The duplicate flag matters — regenerating a role creates a new record rather than replacing the old one.
04
Get the title right before you spend anything
NO ALLOWANCE USED
What comes back: a title recommendation. Internal titles are the most common reason a generated profile comes back thin, and this prompt costs nothing.
05
Scan a whole job family
NO ALLOWANCE USED
What comes back:
how a family is structured in the market, assembled from published roles. Good input for a job architecture conversation before anything is generated.
Three published role profiles, open to anyone.
Real generated output, not screenshots. Open one before you decide whether the prompts below are worth your time.
Frontline clinical role with regulatory and certification requirements.
Open the profile →
Operational leadership role spanning safety, throughput, and people management.
Open the profile →
Technical specialist role where proficiency distinctions carry real consequence.
Open the profile →
Score a job description
WorkforceGPT scores a job description on how cleanly it converts into a structured role profile — not on how well it is written. Each of these uses one assessment, so the prompts below are written to extract everything a single assessment returns.
06
The full diagnostic
USES ONE ASSESSMENT
Score this job description with WorkforceGPT. Show me the overall score, the sub-scores with their weights, which sections came back present versus missing, and every issue ranked by severity. Then explain in plain language what the score means.
[paste the full job description]
What comes back: far more than a number. Asking for the sub-scores and severity-ranked issues is the difference between “we scored 30” and knowing exactly why.
07
Rank the fixes by what they’re worth
USES ONE ASSESSMENT
Score this job description, then rank what to fix by how many points each fix is worth — use the sub-score weights to work out where the headroom actually is. Be specific about information the document doesn’t contain, not about wording.
[paste the full job description]
What comes back: a triage list ordered by impact rather than by the order issues were found. Clarity edits are usually worth far less than they feel like they should be.
08
Compare two versions of the same role
USES TWO ASSESSMENTS
Score both of these job descriptions for the same role and put the results side by side, sub-score by sub-score. Which one gives us more to work with, and what does the weaker one need to catch up?
Version A:
[paste]
Version B:
[paste]
What comes back: a comparison you can take to a stakeholder. Useful when two business units maintain their own version of the same job.
09
Turn the score into something you can send
USES ONE ASSESSMENT
Score this job description, then write three sentences I can send to the hiring manager explaining why we can’t build a defensible role profile from it yet. Neutral tone, no jargon, focused on the information that’s missing rather than on quality.
[paste the full job description]
What comes back: the score translated into something forwardable. The framing matters — a low score is about missing information, not bad writing.
10
Read back a score you already ran
NO ALLOWANCE USED
What comes back:every assessment you’ve run, from the connector and the web app. Retrieving a past score costs nothing — re-running one costs an assessment, so check here first.
Build a role profile
Give WorkforceGPT a title and an organization and it builds the structured role behind it: description, responsibilities, required skills, proficiency expectations, and aligned learning. Generation runs in the background and finishes in minutes, not seconds — and only one runs at a time.
11
Build with your job description as source context
USES ONE GENERATION
[job title]
at
[organization],
using this job description as source context. If the title can’t be matched to labor-market data,
build it from the document instead. Then tell me which parts came from my document and which came from market data.[paste the full job description]
What comes back:a grounded profile plus a provenance read. Start here rather than with a bare title — supplying the document is the difference between a grounded profile and a title-only guess, and it prevents the most common generation failure.
12
The straightforward build
USES ONE GENERATION
[job title]
at
[organization].
Show me the responsibilities and the required skills with their proficiency levels as a table.
What comes back:
a structured profile, not prose. Naming a real organization matters — it anchors the role to a real context rather than a generic title.
13
Focus on proficiency, not the skill list
USES ONE GENERATION
[job title]
at
[organization],
then show me only the proficiency expectations: each skill, the target level for this role,
and what that level looks like in practice. Skip everything else.
What comes back:
the part that makes readiness measurable. A skill list without a target level can’t produce a gap, which is why this view matters more than the skill names.
14
Connect the role to learning
USES ONE GENERATION
[job title]
at
[organization]
and include aligned learning resources from
[Skillsoft or LinkedIn Learning].
Then group the learning by skill and by the proficiency level it helps someone reach.
What comes back:
the role with development attached, organized by target level. Name the catalog explicitly — Skillsoft is the default, so say so if you want LinkedIn Learning.
15
Build one level, sketch the rest
USES ONE GENERATION
[job title]
at
[organization].
Then tell me — without generating anything further — what you would expect to change at the level above and the level below: which skills appear or drop, and where the proficiency targets move.
What comes back:one governed profile plus a leveling sketch. The sketch is the agent reasoning, not generated role data, so treat it as a planning input rather than a standard.
Work with what you have
Reading back roles you’ve already created costs nothing. These prompts get more value out of generations you’ve already spent.
16
Compare two of your own roles
NO ALLOWANCE USED
[role A]
and
[role B]
from WorkforceGPT. Show me the skills they share, the skills unique to each,
and every place the proficiency targets differ.
What comes back:
an overlap and differentiation read — the underlying question behind career pathing and internal mobility.
17
Check a role against someone’s current capability
NO ALLOWANCE USED
[role title].
I’ll describe where someone currently sits on each of those skills. Compare their level to the target
level for the role, skill by skill, and tell me where the real gaps are.
What comes back:
a gap read against an approved target rather than an impression. Describe the person’s levels yourself — WorkforceGPT holds role standards, not employee records.
18
Find inconsistency across your own roles
NO ALLOWANCE USED
similar roles with very different proficiency targets, organization names spelled inconsistently,
or levels that don’t line up. Rank what you find by how much trouble it would cause downstream.
What comes back:
a self-audit. Inconsistency across role profiles is what breaks readiness reporting later, and it is hard to see without reading them all at once.
19
Turn a profile into a review agenda
NO ALLOWANCE USED
[role title]
and turn it into a review agenda for the subject matter expert who has to approve it:
the decisions they need to make, the assumptions worth challenging,
and the questions only they can answer. Keep it to one page.
What comes back:
a prep document for the human approval step. Generation is the fast part; getting an SME to review well is the slow part.
20
Pull one profile in full
NO ALLOWANCE USED
[role title]
in full — description, responsibilities, every skill with its proficiency level, and the learning attached to each. Format it so I can paste it into a document.
What comes back:
the complete profile. Worth knowing you can retrieve any generation as many times as you like without spending another one.
Chain it together
Longer prompts that run several steps in one pass. These are where the connector earns its keep over copy-and-paste.
21
Score, then build from what the score inferred
ONE ASSESSMENT
ONE GENERATION
Score this job description, then tell me what role title and organization you inferred from it. If that title is a better market match than ours, build the role profile using it — and pass the job description in as source context.
[paste the full job description]
What comes back:
a score, a title correction, and a grounded profile in one pass. The assessment identifies the market title for you, which is otherwise a separate guess.
22
Search first, score only if needed
MAY USE ONE ASSESSMENT
[job title]
and tell me what it contains. Two: if there’s no good match, score the job description below. Three: recommend whether we start from the public role or build our own, and explain why. Stop there and wait for my decision.[paste the full job description]
What comes back:
a recommendation, not a fait accompli. Searching first is free, so it goes first — and the explicit stop keeps anything from being spent before you’ve decided.
23
Scope a job architecture project
USES SEVERAL ASSESSMENTS
[number]
roles in
[function].
First search the public library for coverage of this family. Then I’ll paste two representative job descriptions for you to score. Finish with a one-page readiness brief: what our source material can support today, where the gaps are, and what we should fix before generating at scale.
What comes back:
a scoping document, before budget is committed. Two documents is deliberate — scoring every description in a family will exhaust an allowance quickly.
24
Prepare the handoff to governance
NO ALLOWANCE USED
What comes back:
a candid handoff. The connector can draft role content but cannot approve or publish it — this prompt makes the handoff to a human reviewer explicit rather than assumed.
A low score is about missing information, not bad writing.
The score measures how much of a structured role profile your document can supply versus how much would have to be invented. A well-written job description can score poorly, and that is the point.
The diagnostic that saves the most time:
if ambiguity scores high while everything else scores low, your content is fine and the structure is the problem. Fixing sections and headings will move the score further than any amount of rewriting.
What these prompts will not do
- Nothing here publishes or approves anything.
No tool in the connector publishes a role to the public library, approves a profile,
or writes to your job architecture. Publishing is a decision about your public footprint,
so it stays a deliberate action you take yourself. - No employee records.
WorkforceGPT holds role standards and work requirements. For a gap analysis on a person, describe their current levels in the prompt — the connector has no view of individual employees. - No taxonomy construction or market-drift monitoring.
Those remain platform capabilities and are not exposed as agent-callable tools. - No file uploads.
Prompts pass text. Extract from a PDF or Word document before pasting, or use the web app.
Start with prompt 01. It costs nothing.
Create an account, connect it to your agent, and see real published role profiles before you spend a thing.
