Facilitator Guide - Build-a-Skill Session (Non-Technical)
Single source of truth lives elsewhere. The five ways of working are in
../ways-of-working.md- link to them as Way #1..#5, never restate them. The recipe everyone fills in isSKILL-template.md. Practice data is insample-data/, scenarios inscenarios.md, and finished references inskill-examples/.
Overview
You’re running a collaborative showcase, not a competition. ~15 people at café tables of 3–4. By the end, every person walks out with one real, reusable Skill - a one-page recipe they paste into any chatbot to repeat a boring task reliably. No code. No leaderboard. The win is a wall of Skills people actually want to steal from each other.
This session is tool-agnostic. ChatGPT, Gemini, Claude - whatever they brought. The Skill is a paper artifact; the chatbot is interchangeable.
You need two people:
| Role | Who | Does |
|---|---|---|
| Facilitator | Runs the room | Presents, keeps time, drives the guided build on screen, runs the showcase. |
| Skill Scout | Roams | Floats the tables during the independent build, drops praise call-outs, carries the “borrow a brain” tokens, spots the three hat-tips. |
The Scout is your secret weapon. Their whole job is to catch people doing the right thing and say so, loudly, by name.
The arc of the session
| Block | Time | Minutes | What it is |
|---|---|---|---|
| Warm-up | 0:00–0:08 | 8 | Icebreaker + hand out printouts |
| Why + how | 0:08–0:20 | 12 | The pain→promise, walk the template, the 3 habits, the deal |
| Guided build | 0:20–0:42 | 22 | Build SKILL-status-report together, lock-step |
| Independent build | 0:42–1:22 | 40 | Everyone builds their own; Scout roams |
| Showcase | 1:22–1:45 | 23 | Gallery walk + dot-vote, then spotlight 4–5 |
| Close | 1:45–2:00 | 15 | Monday commitments + blockers + word cloud |
Total: 120 minutes. Print the run-of-show below and keep it in front of you.
Run of Show
0:00–0:08 - Warm-up (8 min)
Goal: Loosen the room, surface the pain, get paper into hands.
- As people sit, hand each person two printouts: a one-pager of the five ways of working and a blank
SKILL-template.md. Have spares - people lose the template the second they start writing on it. - Icebreaker, round the table: “Name one boring, repetitive task you wish a robot would just do for you.” 20 seconds each. No laptops open yet.
- As you hear them, jot a few on the flip chart. These become independent-build candidates - you’ll point back to them at 0:42.
Say this: “Hold onto that task. By 11 o’clock you’ll have built a little robot recipe that does it - and so will everyone else, and you’ll get to steal theirs.”
0:08–0:20 - Why + how to build a Skill (12 min)
Goal: Everyone understands what a Skill is, why it beats a clever one-off prompt, and the three rules they’ll follow.
The pain → the promise (2 min). The pain: you write a half-decent prompt, get a half-decent answer, and next week you start from scratch and get something different. The promise: a Skill is the prompt written down once, with the inputs, the output shape, and the guardrails around it - so it works the same every time, for you and for the person next to you.
Walk the 7-section template on screen (6 min). Put SKILL-template.md up and read it top to bottom. The order is the whole point:
| Section | One-line pitch to the room |
|---|---|
| 1. Goal | What real task does this replace? |
| 2. Inputs + PRIVACY CHECK | What you paste in - and what to strip out first |
| 3. Output | What “good” looks like - you write this before the prompt |
| 4. Instructions / Prompt | The exact text, with [SWAPPABLE BITS] in brackets |
| 5. Worked Example | One real, anonymised run - proof it works |
| 6. Guardrails | Privacy + the human check + what it’s not for |
| 7. Reuse notes | Owner, date, where it lives - so it’s not lost |
Land this point: “Notice Output (section 3) comes before the Prompt (section 4). You describe the result before you ask for it. If you can’t describe it, you’re not ready to generate it.” That’s Way #1.
The three habits, as friendly rules (3 min). Don’t lecture the standards - frame them as three things they’ll catch each other doing today:
| Habit | Friendly rule | Way |
|---|---|---|
| Spec-first | ”Say what good looks like before you ask.” | Way #1: Describe before you generate |
| Sign the work | ”AI drafts; a human always checks and owns it.” | Way #2: A human always signs the work |
| Mind the data | ”Treat every prompt like a public postcard.” | Way #3: Mind what you feed it |
The deal for today (1 min). “No competition. No scores. One goal: everyone leaves with one Skill that works, on paper, that someone else would want to use. We help each other get there.”
0:20–0:42 - Guided build, lock-step (22 min)
Goal: Everyone fills in the template with you, building the same Skill - SKILL-status-report - so the first win is locked before they’re on their own. Reference version: skill-examples/SKILL-status-report.md.
Project a blank template. The input is sample-data/sample-meeting-notes.txt - messy Monday call notes. Everyone follows along on their own paper. Go one section at a time; don’t race ahead.
| Min | Section | What you do on screen |
|---|---|---|
| 0:20 | Goal | ”Turn messy weekly call notes into a consistent status update.” Type it. Move on. |
| 0:22 | Inputs + PRIVACY | List the input. Then run the privacy beat below. |
| 0:26 | Output | Define the shape before writing any prompt - a 4-column table: Summary | Decisions | Risks (R/A/G) | Actions (Owner, Due). Max 8 rows. |
| 0:30 | Prompt | Write it together, then narrate one refinement (below). |
| 0:36 | Run it + Oversight beat | Paste, generate, then run the “find one thing wrong” beat (below). |
| 0:40 | Capture | Drop the cleaned-up prompt + the good output into sections 4 and 5. First win locked. |
▸ The 30-second PRIVACY beat (at ~0:23). Pull up the meeting notes and ask: “Before this goes anywhere near a chatbot - what shouldn’t we paste?” Take three hands. The notes contain exactly three things to catch:
- Real names - Maria Gonzalez, Dave → “Person A”, “Person B”
- A vendor name - Brightwave Ltd → “Vendor 1”
- A confidential figure - the 480k budget ceiling, marked confidential → remove it
Say this: “This is Way #3. Public postcard, not a sealed envelope. We anonymise once, in the Inputs section, and it’s handled forever after.” Note the PRIVACY CHECK line already lives inside the template - point at it.
▸ Narrate one prompt refinement (at ~0:32). Write a deliberately thin first prompt - “Summarise these notes.” - and run it (or just describe what it’d give you: a vague paragraph). Then refine out loud:
“That’s a summary, not a status report. Watch - I’ll add the output shape from section 3, and one anti-guessing line: ‘If an owner or date is missing, write TBC - do not guess.’ Same notes, much better result.”
This is the moment they see spec-first pay off in real time. Small change, big jump. (Way #1, and the small-step habit of Way #4.)
▸ The OVERSIGHT beat - “find one thing the AI got wrong” (at ~0:37). Generate the real output. Before anyone copies it, ask the room: “Don’t trust it yet - find me one thing the AI invented or got wrong.” The notes are deliberately ambiguous, so the AI will guess. Likely catches:
- It assigns the vendor chase to a named owner the notes weren’t sure about (“dave picking up the vendor chase i think”).
- It invents a due date the notes never gave.
- It states the budget figure you just told everyone to strip.
Say this: “That’s Way #2. AI drafts, a human signs. You just did the signing. That’s why the template has a Guardrails section - write down the one or two things you must always check.”
End the block by capturing the refined prompt and the corrected output into the template. First win locked - everyone now has a complete, working Skill in their hands as a worked model.
0:42–1:22 - Independent build (40 min)
Goal: Each person builds their own Skill - for a real task they actually do.
Kick-off (2 min). “Now your turn. Pick a task. Two ways in:”
- A shipped scenario from
scenarios.md- ready-made briefs with matching practice data insample-data/(expense triage, request inbox sorting, document-pile summarising, and more). Lowest barrier; pick this if your own task feels fuzzy. - Your own real task - the boring thing from the icebreaker. Point at the flip chart. Highest payoff; what they take back to Monday.
Reassure the mixed-confidence room: “Stuck on a blank page? Take a scenario card. You’ll still leave with a real Skill.”
Same rules as the guided build. Output before prompt. Anonymise in Inputs. Find one thing the AI got wrong before you trust it. Tablemates are allowed - encouraged - to read each other’s drafts.
Skill Scout roams the whole block. Praise specifically and by name (script below). The Scout also carries the tokens:
🎟️ “Borrow a brain” token. Each table gets one token. Spend it to pull the Scout (or a confident person from another table) over for five focused minutes of help - on the prompt, the privacy strip, anything. One per table keeps the Scout free to roam; spending it is a normal, encouraged move, not a sign of failure.
Checkpoints - call these to the whole room:
| Time | Call |
|---|---|
| 0:42 | ”Pick your task. Output section first - describe ‘good’ before you prompt anything. GO.” |
| 1:07 (15-min left) | “Fifteen minutes. You should be running your prompt and checking the output by now - not still polishing the goal.” |
| 1:17 (5-min left) | “Five minutes. Fill in Guardrails and Reuse notes - owner, date, where it’ll live. Then pens down.” |
Watch for the over-builder: someone trying to make one Skill do five jobs. Steer them back - “One task, done well, beats a Swiss Army knife. That’s Way #5 - keep it simple.”
1:22–1:45 - Showcase (23 min)
Goal: Make the wall of Skills visible, let people vote with their feet, then celebrate a handful out loud. Collaborative, not competitive.
Gallery walk + dot-vote (10 min, ~1:22–1:32). Everyone tapes their Skill (or lays it on their table). Hand out three dot stickers each. “Walk the room. Read at least six Skills that aren’t yours. Put a dot on any you’d genuinely reuse - the sticker means ‘I’d actually use this Monday.’” Music on. This is the social heart of the session - let it breathe.
Spotlight 4–5 Skills (13 min, ~1:32–1:45). Pick spotlights from dot counts + Scout’s notes (most dots, plus one or two the Scout flagged for craft even if they’re quiet). ~2–3 min each. Ask the author three quick questions:
- What boring task does this kill?
- What did you strip out for privacy?
- What did you catch the AI getting wrong?
Award the three hat-tips during the spotlights (see Recognition below). Hat-tips are claps and a name on the flip chart - no prizes, no points.
1:45–2:00 - Close (15 min)
Monday commitments (8 min, ~1:45–1:53). Hand out commitment cards. “Write one sentence: the one task you’ll use your Skill on next week, and the day you’ll do it.” Round the room - each person reads theirs aloud. Public commitment is the whole point; don’t skip the read-out. Encourage them to drop the Skill where their team will find it - section 7, Reuse notes. (Way #5.)
Blockers shout-out (4 min, ~1:53–1:57). “What’s the one thing - a policy, a tool, an approval - that would stop you using this on real work?” Capture every answer on a “Blockers” flip chart. Don’t solve them in the room; name them, photograph the board, and route them to whoever owns AI enablement. This is gold for the programme.
Word-cloud close (3 min, ~1:57–2:00). One live word-cloud prompt (Mentimeter/Slido): “One word for how you feel about AI at work now vs. this morning.” Project it, read the big words, thank them, photograph the wall of Skills. Done.
Skill Scout call-out script
The Scout’s job is to catch the right behaviours and praise them specifically and by name - generic “nice work” does nothing; naming the technique teaches the whole table. Keep these on a cue card.
Spec-first (someone wrote the Output section before the prompt - Way #1):
“Priya - you defined the output table before you touched the prompt. That’s exactly it. You’re not hoping for a good answer, you’re specifying one. Table, look at how she did this.”
Anonymising (someone stripped names / vendor / figures before pasting - Way #3):
“Stop - Tom just swapped the real names for Person A and B before pasting, and pulled the budget figure out. Public-postcard rule, textbook. That habit is the difference between safe and sorry.”
Catching an AI error (someone found something invented and fixed it - Way #2):
“Did everyone catch that? Asha spotted the AI made up a due date that was never in the notes - and corrected it. The AI drafts, a human signs. That catch is the most valuable thing that’s happened at this table all hour.”
Spending the token well (no shame, encourage it):
“Smart move pulling me over - that’s what the token’s for. Let’s get this prompt tightened in five.”
Scout tip: carry a sticky pad. When you spot a hat-tip candidate, jot the name and the reason so the facilitator can call it at 1:32 without scrambling.
”If running behind” valve
Time slips in the independent build - that’s fine, it’s where the value is. Protect the build and the close; compress the showcase. Never cut the Monday commitments.
| Pressure | Do this | Saves |
|---|---|---|
| Running 5–8 min over at 1:22 | Compress the gallery walk to 5 min (two dots each, read four Skills not six) | ~5 min |
| Still tight after that | Spotlight 3 Skills, not 5 (~2 min each); keep the three-question format | ~6 min |
| Genuinely jammed | Trim the word cloud to a quick verbal round-the-room word | ~2 min |
| NEVER cut | The independent build, the Monday commitments read-out, the blockers board | - |
If you must choose between gallery and spotlight, keep the gallery - everyone seeing everyone’s work is the collaborative payoff. The spotlight is the bonus.
Recognition - three hat-tips (non-competitive)
No leaderboard, no prizes. Three hat-tips: a round of applause and a name on the flip chart. They reward craft, not speed, so the quiet careful builder gets seen as much as the fast one. Award all three during the spotlight block.
| Hat-tip | Goes to | What it rewards | Anchored in |
|---|---|---|---|
| 🏆 Most Reusable Skill | Highest dot count | Something the room would genuinely use Monday | The whole point of a Skill |
| ✍️ Cleanest Spec | Clearest Output-before-Prompt | Section 3 so precise the AI couldn’t miss | Way #1 |
| 🔍 Sharpest Editor | Best AI-error catch | Spotted something invented and fixed it | Way #2 |
Say this as you give them: “These aren’t prizes - they’re hat-tips. Three habits worth copying: a Skill people reuse, a spec so clear it can’t be missed, and an eye sharp enough to catch the AI making things up. Steal all three.”
Materials checklist
Set the room up 20 minutes before. Café tables of 3–4, no rows.
Printed (per person):
- Five ways of working one-pager - 1 each (+ spares)
- Blank
SKILL-template.md- 2 each (people redraft; have plenty) - Commitment card + pen - 1 each
Printed (shared):
-
scenarios.mdscenario cards - a few per table for those without their own task - Reference
skill-examples/SKILL-status-report.md- 1–2 per table as a finished model -
sample-data/printouts for the guided build and scenarios (meeting notes, expenses, requests, doc pile) - Skill Scout cue card with the call-out script (above)
Room kit:
- Screen / projector for the guided build (blank template + the meeting notes)
- One device per table minimum with a chatbot open - tool-agnostic (ChatGPT / Gemini / Claude, whatever they brought); have 2–3 spare devices logged in
- Wi-Fi password on screen
- Dot stickers - 3 per person
- “Borrow a brain” tokens - 1 per table (printed card or poker chip)
- Masking tape / wall space for the gallery
- Two flip charts: one for icebreaker tasks, one labelled “Blockers”
- Markers, spare pens
- Timer visible on screen
- Mentimeter/Slido word-cloud poll tested in advance
- Music for the gallery walk
- Camera/phone to photograph the wall of Skills and the blockers board
Facilitator mindset
- You are not the expert in the room - you’re the coach. The Skills people build for their own tasks will be better than any example, because they know the task.
- Praise the habit, not the output. A scrappy Skill built spec-first with a clean privacy strip beats a slick one that pasted real client data.
- Protect the quiet builders. The dot-vote and the Cleanest Spec / Sharpest Editor hat-tips exist so it’s not just the loudest table that gets seen.
- The blockers board is a deliverable. Those answers are the real ROI of the day for the wider programme - capture them, route them, follow up.
Companion files: the five ways - ../ways-of-working.md · the template - SKILL-template.md · scenarios - scenarios.md · worked examples - skill-examples/ · practice data - sample-data/.