huashu-speech-coach

intent

this skill applies patrick winston's how-to-speak framework to audit and improve your presentations, videos, or training materials. use it when you need to strengthen your opening, structure key ideas for retention, tighten your slides, or simulate how an audience will actually receive your message. winston's research shows that speaking quality depends far more on learned technique and practice than on innate talent, so this skill systematically applies his core heuristics (cycling, fence-building, verbal punctuation, questioning) to whatever content you're building.

inputs

required:

• your speech outline, video script, slide deck, or training material in text or image form
• the intended audience (e.g. "first-time AI users", "experienced engineers", "product managers")
• the format and duration (e.g. "30-minute in-person workshop", "10-minute video", "keynote with slides")
• any current performance data or feedback (optional: e.g. "video has 20% drop-off at 5min mark", "audience seemed confused here")

context:

• patrick winston's how-to-speak framework (reference: https://youtu.be/Unzc731iCUY)
• understanding that ~20% of your audience is mentally absent at any given moment
• knowledge that human attention is a single channel: if people are on their phones, they cannot hear you

external connections:

• none required (this is a standalone audit and coaching tool)

procedure

step 1: intake and format identification

input: raw material (outline, script, or slides), audience type, format/duration

action: identify what you're working with (opening section, full script, slide deck, etc.) and note the intended context (live audience, video, training session)

output: clear scope definition (e.g. "auditing a 30-minute in-person workshop outline for a mixed-skill audience")

---

step 2: opening audit (empowerment promise)

input: first 60 seconds of your material

action: check whether your opening tells the audience explicitly what they will gain. look for a concrete promise, not a joke or pleasantry. good promise: "in the next 30 minutes you'll learn 3 AI coding tricks you can use today." bad promise: "hi, i'm here to share some thoughts on AI."

output: pass/fail on empowerment promise, plus a rewritten version if needed

---

step 3: cycling audit (3-point repetition)

input: full outline or script

action: identify your 3-5 core concepts. for each one, check that you (a) state what you're about to cover, (b) explain it in depth, (c) recap what you just said. scan for explicit call-outs like "so we've now covered A and B, here's the critical C."

output: map of where cycling happens (or should happen), flagged sections that skip recaps

---

step 4: fence-building audit (boundary clarity)

input: any claims, definitions, or novel ideas in your material

action: look for explicit statements that clarify "this is what i mean" and "this is not what i mean." e.g. "by AI coding i mean end-to-end product development, not just prompting ChatGPT to write hello world."

output: sections with clear boundaries vs. sections that risk confusion, plus suggested "fence" language

---

step 5: verbal punctuation audit (structural signposts)

input: transcript or outline with speaker notes

action: scan for explicit structural markers: "first, second, third", "we've covered A and B, next is the critical C", "now let's shift to...", "to summarize the last section...". count how often you explicitly tell the audience where you are.

output: density map of signposts (note if they're sparse), rewritten sections with added verbal punctuation

---

step 6: questioning and pause audit

input: transcript or notes on interactivity

action: identify every question you ask the audience. check whether you've noted a minimum 7-second pause for live settings, or whether video scripts include a deliberate thought-break (e.g. "pause here and think: why does this happen?"). flag questions that are either too easy (insults the audience) or too hard (no one dares answer).

output: list of existing questions with pause durations, suggestions for new questions, difficulty calibration

---

step 7: slide deck audit (seven sins check)

input: all slides

action: for each slide, check against these violations: (a) reading slides aloud, (b) standing too far from the screen, (c) too much text (font should be 40pt or larger, forcing brevity), (d) decorative clipart or logos, (e) laser pointer usage, (f) complex background patterns, (g) too many points per slide. identify the one slide (if any) that is intentionally complex (hapax legomenon) to demonstrate magnitude or complexity.

output: per-slide violation report, font size audit, visual clutter score

---

step 8: ending audit

input: final 2 minutes of material, last slide

action: check that you do NOT end with a slide that says "thanks", "questions?", or "the end". confirm your last substantive slide lists your core contributions or key takeaways. verify your closing statement salutes (acknowledges the value of) the audience rather than thanking them for their time (which implies they did you a favor). check whether a closing joke exists (fine, because audience is now accustomed to you) and whether it lands naturally.

output: assessment of ending strength, before/after versions if needed

---

step 9: winston stars audit (five-element framework)

input: your core project/idea/concept name

action: evaluate whether you have these five memory anchors: (1) symbol (a visual icon or image), (2) slogan (a catchy phrase), (3) surprise (a counter-intuitive fact), (4) salient idea (what's most memorable), (5) story (how it came to be). note which are missing.

output: audit of all five elements, suggestions for creating missing ones

---

step 10: audience simulation

input: full material, specific audience persona (e.g. "first-time AI user, product manager, skeptical about tools")

action: step into that persona and walk through your content, noting: where would they zone out, where would they get confused, where would they get excited, where would they feel information overload, what would they remember a week later.

output: persona-specific feedback report with marked trouble spots and moments of resonance

---

step 11: scoring and prioritization

input: all audit outputs from steps 1-10

action: score your material on a 10-point scale across five dimensions: (a) opening power (hook + promise strength), (b) structure clarity (cycling, fencing, verbal punctuation all present), (c) interactivity (questions, pauses, audience engagement design), (d) visual aids (slides/props following rules), (e) ending power (contribution-driven close, salute, clear action). generate one overall score plus per-dimension scores.

output: numbered scores 1-10 on each dimension, plus a prioritized list of the 3 highest-impact changes (smallest effort, biggest gain)

decision points

if you are auditing an in-person presentation (vs. video):

• add checks for venue logistics (lighting, mic, room size, seat layout)
• emphasize direct eye contact, movement on stage, body language
• require full attention demand (phones off, laptops closed)
• timing: recommend 11am slot (audience alert but not food-tired)

if you are auditing a video or online talk:

• tighten front 10 seconds (hook must work with no social pressure to stay)
• extend cycling frequency (check every 3-5 minutes for topic callouts)
• note that questions won't get live answers, but pauses still work (they force thinking)
• check for script flow consistency (no awkward jump cuts or pacing breaks)

if you are auditing a training workshop or hands-on session:

• check for interactive breakpoints (where do people practice vs. listen?)
• verify empowerment promise is skill-focused ("you will be able to do X")
• add logistics checks: room size matching expected attendance, materials ready, time for Q&A

if the material has no clear opening:

• coaching mode: assume you are writing the opening from scratch, flag as priority 1

if the material has no clear ending:

• flag as critical; a weak ending erases strong middle content from memory

if cycling is absent or sparse (< 1 recap per core idea):

• recommend adding explicit recaps: "so we've covered X, and here's why it matters..."

if the audience will be fatigued or in a dark room:

• flag all lighting and pacing issues; note that dim lighting is a sleep signal

if this is a video with drop-off data:

• cross-reference drop-off timestamps with content structure; likely causes are missing verbal punctuation or weak cycling at those moments

if the speaker wants to use slides but also wants to memorize a full script:

• note tension: slides work best as prompts, not as crutches; recommend speaker notes instead of full verbatim scripts on slides

output contract

for outline/script audits:

• markdown document with sections matching the 11-step procedure
• each section includes: current state, violations or gaps, rewritten example, severity (high/medium/low)

for slide deck audits:

• annotated slide report (image + annotation for each slide)
• per-slide violation checklist
• summary table showing font sizes, word counts, visual elements

for video script audits:

• timestamped transcript with annotations
• drop-off risk flags
• rewritten sections for weak openings, missing cycling, sparse verbal punctuation

for full presentation audit:

• five dimension scores (opening, structure, interactivity, visuals, ending), each 1-10
• overall score (average of five)
• prioritized action list (top 3 rewrites, estimated effort, expected impact)
• before/after examples for each major change

file location:

• all outputs delivered in single markdown file or annotated PDF, organized by step
• speaker can copy/paste rewritten sections directly into their material

outcome signal

you know the skill worked when:

1. opening: you have a single-sentence empowerment promise that tells the audience what they will gain, not a joke or greeting

2. structure: you can identify 3-5 core ideas and point to at least 2 moments in your material where each idea is cycled (stated, explained, recapped)

3. clarity: for each novel claim, you can point to a sentence that says "this is what i mean" or "this is not what i mean" (fence-building)

4. signposts: scanning your transcript, you see "first/second/third", "now let's move to", "to recap", or similar verbal punctuation at least every 5-10 minutes

5. engagement: you have at least 2-3 questions in your material, each with a noted pause (7+ seconds for live, deliberate break for video)

6. slides (if applicable): no slide has more than 5-6 bullet points, no font smaller than 40pt, no decorative elements, no laser pointer notes

7. ending: your final slide lists key takeaways or contributions (not "thanks"), and your closing statement salutes the audience ("you came here because you care about X" not "thanks for your time")

8. memory anchors: you can articulate all five winston stars for your core idea (symbol, slogan, surprise, salient idea, story), even if some need refinement

9. persona feedback: when you walk through your material as a specific audience member, you can identify 2-3 moments of clarity and 1-2 moments of confusion, and you have rewrites for the confusion points

10. scores: your audit produces a 5-dimension scorecard and a ranked list of 3 highest-impact changes you can implement without major restructuring

the skill succeeds when your revised material increases retention (measured by audience recall a week later, or video completion rates), reduces drop-off at specific timestamps, and produces feedback comments like "i knew exactly what you were asking for" or "i remember your key idea because of X symbol/slogan".

---

created by alchaincyf. based on patrick winston's how-to-speak framework (mit, 2019). curated for creators, speakers, educators, and presenters building audiences or teams.