← back to landing
Workshop · Session 1 of 2

Structured
Prompt Engineering

Move from "the AI gave me a good answer" to "my team can run my prompt without me."

Mon 04/05/2026 · 09:30–12:30 3 hours · 10–12 attendees English · WebMD ONE · TINYpulse
NEXT: smile, breathe. "Good morning. Three hours. One job. By 12:30 each of you walks out with one prompt your team can run without you." Then announce: open the Coach link from the email — keep it open all morning.
Before we go any further · let me name the relationship
Tony Bùi
Tony
CEO · Chief of Everything · Vibery Product Studio
You don't need to remember me. Just care about whether what I show you helps.
Múa rìu qua mắt thợ.
"Don't tell a bird how to fly."
I'm not a product designer.
I'm an AI orchestrator.
You bring the expertise. I bring the orchestration.
3 hours · 1 outcome · here's the shape

Three acts.

01
Frame
Learn the moves.
~45 min
♥ the heart
02
Practice
YOUR real task. Coach in hand.
~80 min
03
Ship
What you take home.
~20 min

Detailed times are in your handout. Most of today is your hands on your work — not me talking.

21 in the room · 5 sent tasks ahead · these are real

Who's here

I got 5 responses. From those 5, I patternized everything into 5 buttons that I hope cover most of your work.

Today we go back to the foundation — every skill, every agent sits on a prompt that works.

Product Owner
Synthesize user interviews → talking points
Sr. Designer
Connect-the-dots across fragmented research
Product Owner
Stories + acceptance criteria + ticket notes
Product Owner
PRD → YouTrack tickets + timeline
UX Manager
Direct-report coaching, status, growth path
The questions you may already be asking

"Why do we need this?"

Three honest questions. Each one has a yes — and a but.

Through-line — prompt practice IS delegation practice. Models change. Delegation doesn't.

5 roles in this room · 5 different weeks · same 5 verbs

5 roles. Same back room.

Five different doors labeled Product Owner, BA, Designer, UX Manager, Researcher — each leading via curved arrows to the same back room where 5 verbs are written on the wall: draft, synthesize, review, critique, translate. Moleskine notebook hand-drawn style.
5 roles in this room → same back room of work.

Click your role. Watch the verbs light up.

draftnew content synthesizemany → one reviewscore vs rubric critiquefind what breaks translateformat / audience
Different role. Different week. Same 5 verbs. We name them today.
Live demo · same task · two prompts

Lucky prompt vs team asset

Same task. Same model. Different prompt structure. The gap between the outputs IS the workshop.

Prompt #1 · lucky
Turn this PRD into YouTrack tickets.

[paste PRD section here]
WHAT AI WILL DO
Invent priorities · skip compliance flags · paraphrase AC · ship a generic blob you rewrite by hand
Prompt #2 · same task · RCTCO
Role: senior PO on a wellness platform that ships to YouTrack.
Context: WebMD ONE · member consent · HIPAA · 2-week sprints.
Task: convert each user story below into one YouTrack ticket.
Constraints:
  - preserve every AC verbatim
  - mark unknowns [TBC: confirm with PM]
  - flag PHI lines [COMPLIANCE — HIPAA]
Output: markdown table —
  | # | Title | Description | AC (G/W/T) | Priority | Labels | Flags |
WHAT AI WILL DO
Real table · [TBC] where unknown · [COMPLIANCE] tags on PHI · ready to paste
AC = acceptance criteria · G/W/T = Given/When/Then · PHI = protected health info · DoD = definition of done

Same model. Same task. Different prompt. That gap is today.

Module 1a · Watch one prompt grow

Lucky 1-liner → RCTCO asset

Click each step. Watch the prompt grow. Five clicks separate "lucky" from "asset."

Survey example · live 
Write user stories for our consent feature.+RYou are a senior Business Analyst reviewing a wellness-platform spec.+CProduct: [PRODUCT] member portal. Users: [USER TYPE]. Regulations: HIPAA + [REGION PDPL]. Source: [SPEC LINK §X.Y].+TDraft [N] user stories with acceptance criteria for the [FEATURE] flow.+CnDo NOT invent regulations. Do NOT paraphrase AC — verbatim only. Cite spec §X.Y per story. Flag PHI lines [COMPLIANCE — HIPAA]. Mark unknowns [TBC: confirm with PM]. Mark guesses [ASSUMPTION].+OMarkdown table. Columns: Story | AC (Given/When/Then) | Spec ref | Flags. One row per story. Each row independently reviewable.
✓ Passes 5 tests · ready for your team
Click in order →

Same task. Five clicks. Lucky → asset.
Coach does these 5 clicks for you in 30 sec — but you had to see the move once.

Beyond RCTCO · the pattern underneath

All frameworks rhyme

Different acronyms. Same five questions. Prompting is delegating. Good delegation asks the same questions — whether the assignee is human, agent, or AI.

Delegation question
RCTCO
CRISPE
RTF
COSTAR
Who am I asking?
Role
Capacity
—
Style
What do they need to know?
Context
Insight
—
Context · Audience
What is the job?
Task
Statement
Task
Objective
What are the rules?
Constraints
Personality
—
Tone
What does done look like?
Output
Experiment
Format
Response
Reproducibility
Boundedness
Testability
Reusability
Auditability

Stop arguing about frameworks. Learn the delegation pattern. The acronym is just the wrapper.

The zoom-out frame

5 tests every prompt must pass

Skip one slot of RCTCO → fail one test. Each slot defends one test.

01
Reproducibility

Can my teammate run this and get the same shape — without asking me a question?

R + C
02
Boundedness

Did the AI stay inside what I asked, or did it invent things?

Constraints
03
Testability

Can I check if the output is correct, or is it just vibes?

Output format
04
Reusability

Can I run this same prompt next sprint, only changing 2–3 inputs?

All 5 stable
05
Auditability

If a regulator asks "where did this come from" — can I show the trail in 60 seconds?

Constraints + Output

For healthtech / HRtech — Auditability is the test that breaks AI use. Most teams skip it and find out the hard way.

Module 1b · the iteration loop

The 3-turn iteration loop

Most people throw away output and start over. Wrong move. Three turns get you 10× the result for 2× the time.

Turn 1
RCTCO prompt
Full 5-slot prompt → first output (output A)
→
Turn 2
Self-critique
"Critique your output against these 3 tests. List 3 weaknesses."
→
Turn 3
Rewrite
"Now rewrite addressing those weaknesses." → output B

Turn 2 is just the AI grading itself against the 5 tests. Pick 2–3 tests per turn 2 — depends on your domain.

5 named patterns

The 5 shapes underneath

5 different tasks in this room. 5 patterns — same shape, different content. Name the pattern → know the prompt.

DRAFTER

Draft new content

PRDs, user stories, copy. Mark every assumption [TBC].

stories + AC · ticket drafts
SYNTHESIZER

Synthesize many inputs

Cluster interviews / docs / threads. Cite source IDs. No fake quotes.

interview synthesis · research dots
REVIEWER

Score against criteria

Existing work + your rubric. Severity-tagged findings, row by row.

spec review · DR coaching
CRITIC

Find what could go wrong

A plan, no rubric. Adversarial — edge cases, failure modes, missing scenarios.

pre-launch stress-tests
TRANSLATOR

Translate format / audience

PRD → tickets. Tech → patient-facing. Preserve meaning, change shape.

PRD → YouTrack tickets

Reviewer vs Critic — both find problems, different jobs. Reviewer asks "does this meet the bar?" against an existing rubric. Critic asks "what could break?" against an unproven plan. Different prompts, different patterns.

Recipe card 5-pattern structures + worked examples
Coach install · 60 sec to 3 min · pick your path

Open your assistant

Coach is your workshop tool for the next 3 hours. Three install paths — pick the one that fits your Claude account.

RECOMMENDED · 60 SEC

1 · Skill

Needs: Claude Pro / Max / Team

Upload the skill .zip to Claude. Toggle on. Done.

ANY PLAN · 3 MIN

2 · Project

Needs: any account (Free OK)

Build a Claude Project from our zip. We walk you through.

FALLBACK · 30 SEC

3 · Paste-block

Needs: any chat (incl. logged out)

Copy a big block. Paste as message #1 in any new chat.

Open this link → pick your path
vibery-studio.github.io/workshop-0504
When you're in, type "in" in chat. Stuck? Type "help".
10:45 — 11:00
Break · 15 minutes · Come back at 11:00

After the break — your turn. 80 minutes on your real task.

Module 2 · 80 minutes · You + Coach + me

Your turn — with your assistant

21 of you. 21 Prompt Coaches running in parallel. Coach scores. I steer.

MODE 1 · BUILD

You paste a bare task

"Help me write a prompt that turns my PRD into YouTrack tickets." Coach asks 5 RCTCO questions tailored to your task. You answer. Coach assembles + audits.

MODE 2 · AUDIT

You paste a drafted prompt

"Score this prompt: You are a senior PO..." Coach scores 1–5 on each of 5 tests. Total /25. Names 2 highest-impact fixes.

MODE 3 · DIAGNOSE

You paste an output that disappointed you

"AI gave me this and it's wrong: ..." Coach traces failure to the missing RCTCO slot. Gives the exact constraint to add.

MODE 4 · BUILD-CARD

You paste raw team notes

"Format this as a context card: vocab — member not user; DoD — AC in G/W/T..." Coach formats your raw notes into a saveable Context Card markdown.

MODE 5 · PERSONALIZE

You paste a finished Context Card

"TEAM CONTEXT — Vocab: member not user. DoD: AC in G/W/T..." Coach loads your team standards. Every later prompt is scored against them too.

Warm-up · 5 minutes
Paste anything into Coach now — bare task, draft, or bad output. See which mode triggers.
Type "done" in chat when Coach replies. Then we start Round 1.

Coach never runs your prompt for you. It coaches the prompt. You run it in another tab. That separation is the discipline.

Module 2 · Round 1 · 15 minutes

Pick your pattern. Build with Coach.

YOUR MOVE · 15 MIN
  1. Pick the pattern that fits your real task — draft / synthesize / review / critique / translate
  2. Open Coach. Paste your bare task in 1-2 lines.
  3. Coach asks 5 RCTCO questions. Answer them honestly.
  4. Coach assembles your prompt.
  5. Save as artifact — paste in chat: "put this prompt in an artifact called r1-prompt.md so I can edit it across rounds."
  6. Type "R1 done · artifact saved" in chat.
If you're stuck
No real task on your plate? Use this: "draft 5 user stories with AC for [your current feature], must cite spec, output as table." Same shape. Generic enough.
⏱ 15:00 — Coach BUILDS your prompt with you
Goal: a working RCTCO prompt · not perfect · working
Module 2 · Round 2 · 20 minutes

Audit. Score /25. Fix.

YOUR MOVE · 20 MIN
  1. Input: your r1-prompt.md artifact from R1 (or paste from chat).
  2. Ask Coach: "audit r1-prompt against the 5 tests."
  3. Coach scores /25 + names 2 fixes + emits a FIXED PROMPT block.
  4. Ask: "update the r1-prompt artifact with the FIXED PROMPT." Coach overwrites in place.
  5. Re-audit. Repeat until score is 18+/25.
  6. Type "R2 [score] · artifact at v2" in chat.
Scoring bar
< 14/25 · personal trick — only you can run it
14-17/25 · works in your hands · breaks in others'
18+/25 · team asset · ship Monday
22+/25 · context-injected (Round 3)
⏱ 20:00 — Coach AUDITS · you fix · re-audit
Goal: 18+/25 · ship-ready
Module 2 · Round 3 · 25 minutes · the team-asset round

Inject your team. Score → 22+.

YOUR MOVE · 25 MIN
  1. Input: your r1-prompt.md artifact from R2 (now at 18+/25).
  2. Open a fresh chat with Coach (clean context for R3).
  3. Build your Context Card (right →) — vocab, DoD, 1 example. Two paths:
    • Solo: use the right-column reference + deep-dive link to fill by hand.
    • With Coach: dump raw notes, ask "format this as a Context Card I can save".
  4. Ask Coach: "put this Context Card in an artifact called team-context-card.md."
  5. Paste your R2 prompt + ask: "score r1-prompt with team-context-card applied." → goal 22+/25.
  6. Ask Coach to update r1-prompt with any final fixes.
  7. You walk out with TWO artifacts — your prompt + your Context Card. Both reusable forever.
  8. Type "R3 [score] · 2 artifacts" in chat.
VOCAB
"member" not "user" · "engagement" not "DAU" · "claim handoff" not "API call"
DoD
AC in G/W/T · [COMPLIANCE — HIPAA] if PHI · 2-lead review · points TBC if unknown
EXAMPLE
One real ticket / story / spec your team shipped. Paste verbatim. AI mirrors style.
⏱ 25:00 — Context Card → re-audit → save
Goal: 22+/25 · YOUR team's prompt
Deep-dive Context Card template + vocab/DoD/example worked examples
Module 2 · close · 10 minutes · 2 volunteers

Share your prompt. Room learns from each other.

VOL 1 · 5 MIN

High score (22+)

Paste R3 prompt + score in chat. Walk us through your 3 injects. Room reads. One Q from the room.

VOL 2 · 5 MIN

"Mine is below 18 and I want help"

Brave volunteer. Tony pastes their prompt into Coach on screen, audits live. Room watches the move.

Watching someone else's prompt audited — you see your own gaps in their prompt. That's why we share, not lecture.

The meta-reveal · Coach IS a Project

You've been using a Claude Project all morning

Prompt Coach = a Claude Project. System prompt + knowledge files + share link. The same pattern works for ANY repeated workflow on your team.

Without a Project

Paste 5 slots, every chat

  • Re-paste Role + Context every time
  • Teammate forgets one slot → different output
  • Knowledge docs not attached → AI guesses
  • No shared baseline
Like Prompt Coach

Slots locked. Inputs vary.

  • Role + Constraints + Output = system prompt (locked)
  • Knowledge files = your team's vocabulary, templates, regs
  • Teammates open one link, get the same shape every time
  • Reproducibility test passes by default

Coach was your training wheels. Build your own Project this week — for PRD review, for pulse-survey synthesis, for ticket translation. One per recurring task.

From 1 person → your team

Where good prompts live

A Project belongs to one team. A repo belongs to your whole company. Same four properties your code and specs already have — applied to prompts.

VERSIONED

Versioned

Today: git history for code & specs.
Prompts: prompts/reviewer.md committed, semver-tagged.

REVIEWABLE

Reviewable

Today: code review & PRs.
Prompts: "What failure mode does this defend?" PR template.

DISCOVERABLE

Discoverable

Today: README + folder structure.
Prompts: prompts/README.md + folder per pattern.

RELEASABLE

Releasable

Today: tagged versions.
Prompts: v1.0 before audit, v1.1 after harvest.

Plain-Git is durable. SaaS comes when you outgrow markdown. Start where your team already is.

Deep-dive Folder structure · PR template · versioning · eval tooling tiers
From today → Session 2 (May 11)

The ladder

Same pattern, four rungs. Each rung tests harder than the last.

LEVEL 1Today · 09:30

Prompt

RCTCO + 3-turn loop. Passes 2-3 of 5 tests on its own.

Lives in your hands
LEVEL 2Today · 11:00

Project

Same prompt + locked knowledge + shareable. Passes Reproducibility + Reusability.

Lives in one team
↑ UPGRADE: Cowork (desktop) — same Skill, opens local files
LEVEL 3soon · for tech leads

Repo

Your team's prompts/ folder. Versioned, reviewable, owned. Where prompts become infrastructure, not experiments.

Lives in shared git
LEVEL 4May 11 · Session 2

Skill

Claude Code triggers your prompt automatically on keywords. Passes all 5 tests.

Lives on autopilot

By Repo, your prompts survive without you.
By Skill, they run without you.

Deep-dive Project recipe · Cowork · hybrid team pattern · the full ladder
What you take home · the contract

4 things in your pocket. Forever.

01 · COACH

Prompt Coach

Same Coach you used today. Open Monday, paste your next real task. Yours forever.

link in email
02 · CONTEXT CARD

YOUR team's voice

The card you filled in Round 3. Vocab + DoD + 1 example. Paste before any prompt — instant team-asset quality.

save it forever
03 · CHEAT SHEET

1-page reference

5 patterns · RCTCO · 5 tests · 3-turn loop. Print it. Pin it. Pass it to a teammate.

PDF in email
04 · REPO IDEA

For your tech lead

Bring prompts/ folder to your tech lead Monday. The team-asset move is real.

in email

If you leave without all four — the session failed. That is my contract with you.

Homework for Session 2 · Mon 11/05
Bring one repetitive task in your weekly workflow.
We turn Coach into a Claude Code skill that runs it on autopilot.
Comic recap Today's session in 9 frames Notebook · 15 illustrations