Copilot TrainingCPS HR Training Center · Tue Oct 20, 2026
Your progress
0%

Frameworks and toolkit

Your Copilot toolkit

The frameworks from class, in one place, plus a 30-day action plan you can fill in and keep.

Prompt ingredients and source checks

Developed by Dr. Frank Olmos. Start with C.H.A.T., the four required parts. Each layer of P.R.O.M.P.T. is added one lab at a time. Guardrails stay separate: what Copilot must not infer, how you will check the source, and who decides.

The C.H.A.T.P.R.O.M.P.T. Pyramid

Build from the bottom up. Every prompt needs the four foundation blocks. Add the upper layers as the task gets harder.

    • ContextWhat is going on
    • How come / Why?Why you need it
    • Assign taskDefine the outcome you want
    • Type of outputThe shape you want: list, table, memo

    Foundational building blocksC.H.A.T.: use in every prompt

    • PersonaWho Copilot should act as
    • Reference examplesSupplied files or samples to use and cite
    • OptimizeImprove the answer in rounds (iterative refinement)

    Intermediate strategiesAdd when the task needs more

    • ModularizeSplit the work into smaller subtasks
    • ProgressionAsk for step-by-step reasoning (chain of thought)

    Advanced techniquesFor multi-step work

    • TranscendenceMastery and innovation: the next insight after the checked draft

    Expert levelWhere practice leads

© 2026 Dr. Frank Olmos. The C.H.A.T.P.R.O.M.P.T. framework and pyramid were developed by Dr. Frank Olmos, DFO HR Consulting.

Contextwhat is going on
How comewhy you need it
Assign taskwhat to do
Type of outputthe shape you want
Personalayerwho Copilot should act as
Referenceslayerthe supplied files Copilot should use and cite
Optimizelayermake the response useful for its purpose and reader
Modularizelayerbreak the work into clear, reviewable steps
Progressionlayerthe order the response should follow
Transcendencelayerexpert level: the next insight or question after the checked draft

Try it in the prompt builder

Five Checks reference

Before using an AI answer, check where it came from, whether it is correct, what might be missing, and who has authority to decide.

SourceCan I trace it?
VerificationDid I open the names, dates, and numbers?
ContextWhat is missing?
ResilienceWould it still be accurate if someone read it without the chat, or hit an exception?
Authorized OwnerIs this a requirement or a suggestion, and who is authorized to decide?

Worked example: Copilot says the deadline is 7 days. An email announced 7; the approved procedure says 5. Confirm which one controls before you communicate it.

Five labels stay separate: Decision (agreed by someone with authority) · Proposal (suggested, not yet agreed) · Owner (a person who accepted the task) · Target (a hoped-for date, not a commitment) · Open question (something the source does not answer yet).

Download the Five Checks (.docx)

Four ways to work with Copilot, and how agents fail

Choose the tool

Copilot Chat: general help. Researcher and Analyst: built-in agents for research and data. Agent Builder: your instructions and documents; answers still need checking. Copilot Studio: configurable agents and workflows. Action agents need a named owner and a person's approval for anything that matters.

Agents fail when they...

1. Answer from outside their sources. 2. Blend documents that disagree. 3. State a rule they were never given. 4. Act without approval.

Governance checklist: responsible-use policy, data protection, human decision authority, quality and fairness, transparency, and an escalation path.

Your 30-day Copilot action plan

For each line, name yourself as owner, a dated action, and a measure. A wish: "Use Copilot more." A plan: "I will use the Five Checks to review Friday's management briefing by October 23, 2026." Your notes save in this browser only.