← All simulators
AI skills · Prompt writing
From SOP to job aid
A hands-on lab for building practical AI skills. Write a prompt, run it against a real-world messy procedure, and get coaching on both the prompt and what it produced.
Work sample by Allan DeYong, built on the OpenAI API. All companies, people, and documents are fictional.
The task
- Your task
- Write a prompt that turns the Dock 2 receiving procedure into a one-page job aid for a new receiving clerk on their first solo shift.
- How it works
- The procedure is attached to your prompt automatically. Run your prompt, read the result, revise, and run again. Ask for coaching whenever you’re ready.
- Watch for
- The procedure is realistic: run-on sentences, buried safety steps, and one form that was discontinued.
The procedure (attached automatically)
RECEIVING INBOUND FREIGHT - DOCK 2 (rev 4, owner: J. Marsh, Ops) When a trailer is assigned to Dock 2 the receiver should first confirm the trailer number against the appointment in the WMS, if it doesn't match do not open the doors and call the yard lead (ext 214). Wheel chocks and the dock lock must be engaged before anyone enters the trailer, this is non-negotiable and was the subject of the March incident. Check the seal number vs the BOL, note it on the BOL and photograph it with the RF gun camera. If the seal is broken or missing stop and notify the shift supervisor, don't unload. Unloading: use the electric pallet jack for floor-loaded pallets, forklift only if you are certified (badge shows the green sticker). Count pallets and compare to the BOL piece count. Any damage (crushed corners, wet cartons, leaning pallets over 2 inches) gets photographed before it moves, and written on the BOL as "subject to inspection - damage noted" and the driver signs next to it. Previously we used OS&D forms but those were discontinued in rev 3, just use the BOL notation now. Scan each pallet's LPN label into the WMS receiving screen. If a pallet has no label print one from the dock 2 printer (Zebra, it jams if you use the old roll - use the 4x6 roll only). Overage/shortage: if counts don't match after recount, receive what is there and flag the PO in WMS with reason code SH (short) or OV (over). Hot POs (shown in red in WMS) go straight to staging lane A, everything else to lanes B-D by zone. Driver paperwork: sign the BOL only after counts and damage notes are complete, give the driver the pink copy. Release the dock lock and remove chocks only after the driver is back in the cab and you get a thumbs-up. Sweep the trailer floor of debris (shrink wrap, broken pallet boards) before release. Typical time to receive a full 26-pallet trailer: 45-60 minutes. New receivers shadow for 3 shifts before receiving alone.
Your prompt
Write it the way you’d write any work prompt. You don’t need to paste the procedure.
Output
How it’s built
Scenario
A realistic, messy procedure with buried safety steps, escalation points, and one discontinued form. A vague prompt produces a confident job aid with mistakes in it.
Rubric
Four criteria score the prompt itself; the fifth checks the output against the source for accuracy, missing safety steps, and invented content.
Why it matters
Prompting is a writing skill. People improve fastest when they can see what their own prompt produced and why.
The rubric
| Criterion | What strong looks like (4) |
|---|---|
| Audience and purpose | Names who will use the job aid and when: a new clerk, on the dock, first solo shift. |
| Clear task | States exactly what to produce and what matters most, such as step order and stop-and-escalate points. |
| Format and length | Specifies a usable structure: numbered steps, a short “stop and call” box, one page, plain language. |
| Guardrails | Tells the model to stay within the procedure, leave out discontinued items, and flag anything unclear. |
| Output quality | The result is accurate, keeps every safety and escalation step, drops the old form, and is usable at a glance. |
Facilitator guide
- Before: Have learners write a first prompt in under two minutes, the way they would at work.
- Compare: Run the quick prompt, then revise once using the coaching. Put the two outputs side by side.
- Ask, don’t tell: “What did the model get wrong, and which sentence in your prompt would have prevented it?”
- Build a habit: Have learners save their best prompt as a template for their own procedures.
- Carry it to real work: Pick one real procedure from their area and run the same exercise on it.