Prompt like an Operator

Before you start

Prerequisite: distinguish a source fact from a generated claim. You will write a bounded prompt, define checks before drafting and revise against an observed failure. The Riley facts below support an offline, no-cost route with paper and supplied drafts.

A short question can be useful. More detail helps when it supplies relevant facts and criteria; extra text alone does not improve a result. Prompt quality is one part of the workflow alongside model capability, context, evidence, tools and permissions.

What you will be able to do

• Write Context, Task, Standards and Evidence for a bounded draft.
• Use examples without importing their unsupported facts.
• Revise one prompt variable against a specific observed failure.

Part 1 What an Operator is.

In this curriculum, an Operator is a highly trained person whose work demands strict adherence to correct process. Judgment under pressure. Clear-eyed honesty about what they can and can’t do. The term shows up across the military, intelligence, special operations, federal law enforcement, aviation, emergency response, and high-stakes civilian work.

For this curriculum, the Operator practice is to prepare inputs and acceptance criteria before use. They give the system what it needs, in the order it needs it, with the constraints that matter. They read the output. They judge it. They iterate.

The useful habit is deliberate input, critical reading and accountable revision.

Part 2 CTSE makes the task and its checks explicit

• Context. Who you are, what’s going on, what’s already happened.
• Task. What you want the AI to produce. Use a verb. Be specific.
• Standards. Format, length, audience, tone, what to include, what to avoid.
• Evidence. Supplied source IDs, claim checks, calculations or tests, and how to handle missing support.

This is CTSE: Context, Task, Standards, Evidence. Earlier CTS examples remain useful; add Evidence to make acceptance inspectable.

This structure makes instructions easier to evaluate. It does not guarantee correctness or replace tool permissions.

Part 3 Bad prompt vs Operator prompt.

Bad prompt

Help me write a cover letter for the project manager job at Microsoft.

The request supplies little evidence about the applicant or role. A useful response should ask for the missing facts rather than invent them.

Operator prompt

FICTIONAL TRAINING EXAMPLE
CONTEXT: The applicant is a separating Air Force captain (O-3, 8 years, logistics
officer). I'm applying for a Program Manager role on Microsoft's
Federal Cloud team. Here's the job description: [paste]. Here are
my three biggest accomplishments: [list]. Here's a cover letter I
wrote for a previous application: [paste].

TASK: Write a cover letter — one page, three paragraphs.

STANDARDS: Open with a hook about Microsoft's federal mission,
not "I am writing to apply for." Bridge my military experience to
the role using plain language, no military acronyms. Close with a
specific question I'd ask the hiring manager. Tone: confident,
direct, not formal.

Add an EVIDENCE line: use only supplied accomplishments and posting text; mark gaps; verify each factual claim. Neither example establishes hiring success.

Components, broken down

• Context: who you are and what’s happening. The AI doesn’t know any of this. Tell it.
• Task: what output you want. Use a verb. “Write,” “summarize,” “compare,” “draft,” “outline,” “critique.” Be specific about the deliverable.
• Standards: the details that make the output usable. Length, format, audience, tone. What to include. What to avoid. Reference examples when you have any.

Part 4 One more move: give it an example.

If you have a sample of what good looks like, paste it in. “Here’s a cover letter I liked. Match this tone but write it for the job below.” In the literature, one example is “one-shot” prompting and several is “few-shot.” You don’t need the literature. You need the move.

Examples can clarify a format or tone, but they can also bias the model toward irrelevant facts. Label the example as a style reference and prohibit copying its accomplishments. Check whether the example actually improves the target task.

Substantive review: 2026-10-02. Next, use Lesson 04 to audit claim support independently of the draft.