agentleFS
cleysonunez6327cleysonunez6327/videos-demo/skills/testimonial/SKILL.md

testimonial

Skill0 starsChanged 3 months agoMITView on GitHub

Build customer testimonial and case-study demo videos — first-person narration over a real screen recording of the product doing what the customer says it does. Use for social proof, customer stories, case studies, before/after workflow comparisons and result-led product videos. Requires a provenance declaration before rendering. Triggers on testimonial, case study, social proof, customer story, success story, caso de éxito, testimonio, prueba social, historia de cliente.

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 1

    1. igoroliveirg/claude-copy★ 9same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repositoryOne other file it gives its agents.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 5 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not restate any of it here.

    line 36
  2. Very specific:

    1. Establish provenance before anything else: Ask the user whose words these are.

    line 59
  3. Distinctive:

    Record the surface the customer actually touches: Not the marketing site. If the story is told by a buyer, find the buyer-facing page — the …

    line 71
  4. Distinctive:

    The narration may not outrun the screen: Write only what the recording will actually show.

    line 84
  5. Distinctive:

    6. Render and check the disclosure is legible: Not just present — legible at 360p on a phone, which is where the video will actually be …

    line 96
  6. Very specific:

    Shoot vertical when the customer is on a phone: If the story happens at a counter, in a queue or anywhere the person is holding a device, …

    line 116
  7. Distinctive:

    No split screen: Before and after run one after the other, never side by side.

    line 151

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

testimonial

A testimonial in ndemo is not a talking head. There is no camera, no face and no set — there is a browser. So the format that plays to that strength is different from the one a video model would produce:

The product doing exactly what the customer says it does, narrated in the first person.

The viewer does not watch someone claim the tool saved them eleven hours. They watch the eleven hours being saved, while a voice explains what the work used to look like. The screen is real and checkable; only the voice is synthesised, and audioFile removes even that once it lands.

Read first

This skill decides what is told and in what order. It delegates how to ndemo:

  • ../ndemo/SKILL.md — the execution workflow: open the browser, author actions from the accessibility tree, test each segment, render.
  • ../ndemo/references/narrative.md — the base doctrine: the retention clock, leading with proof, how to write narration, publishing formats. Do not restate any of it here; this skill only adds what is specific to a testimonial.

Then read, in this directory:

  • references/provenance.md — read before writing narration. Which route applies decides what you are allowed to say.
  • references/formats.md — the four formats that work on a screen.
  • references/voice-casting.md — which voice belongs to whom.

When this skill applies, and when ndemo does

Both skills trigger on video requests, so the line matters:

The video is told from…Skill
the customer's voice — first person, past tensetestimonial
the product's voice — features, capabilities, a tourndemo

"Show what our app does" is ndemo. "Show how a customer uses it" is this.

Workflow

1. Establish provenance before anything else. Ask the user whose words these are. Do not start writing narration and sort the attribution out later — the route changes what the narration may claim. See references/provenance.md.

2. Pick a format. references/formats.md has four. For a first testimonial, Workflow Proof is usually right.

3. Find the moment worth proving. A testimonial needs one concrete before/after, not a feature list. Ask: what did this person do on a Monday morning that they no longer do? That is the video.

Record the surface the customer actually touches. Not the marketing site. If the story is told by a buyer, find the buyer-facing page — the public checkout, the shared link, the client portal — and record that; many products have one that needs no login at all. A first-person narration running over the vendor's own landing page is the failure mode this format falls into most often, and a viewer spots it within seconds: the voice claims to be using the product while the screen shows a brochure.

4. Write the narration in the first person. Past tense for the pain, present for the relief. Be specific — "three CSVs every Monday" lands, "improved my workflow" does not. Never write marketing register into a customer's mouth; if it sounds like an ad, it performs like one.

The narration may not outrun the screen. Write only what the recording will actually show. The temptation appears wherever the last step costs real money or real consequences — a payment nobody is going to send, an email nobody is going to receive, an order nobody is going to place. The screen stops at "waiting"; if the voice says "and it's paid", the subtitle is contradicting the picture underneath it, and that is the same defect as narrating over a landing page. End on the gesture instead — "I scan the code and confirm" — which is true, and let the interface speak for the rest.

5. Author the actions with ndemo. Same loop as any other playbook: open the browser, read page-state, write actions, test each segment.

6. Render and check the disclosure is legible. Not just present — legible at 360p on a phone, which is where the video will actually be watched.

Settings that differ from a product demo

tts:
  voice: eve          # never `sal` — that is the product's voice
  speed: 0.96         # slightly under 1.0 reads as a person, not an announcer

app:
  zoom: 1.35          # a product demo uses 1.25; here the detail IS the proof

music:
  volume: 0.10        # the customer's voice leads; the bed stays under it

subtitles:
  burn: true          # feeds autoplay muted, and a testimonial is words

Shoot vertical when the customer is on a phone. If the story happens at a counter, in a queue or anywhere the person is holding a device, set the viewport to 1080x1920. The page renders its own responsive layout, so nothing is cropped, and the frame itself says "this is what I saw on my phone" before the narration has to. Raise zoom until the content fills the height — a page that centres a narrow card leaves most of a vertical frame empty at zoom: 1.0.

Where the disclosure goes

On the illustrative route the disclosure is only worth having if it can be read. titleCard renders it two different ways, and the difference matters:

  • With a stat, the title and subtitle are drawn together on one small grey line under the number. A disclosure placed there is unreadable once the video is scaled to a feed.
  • The stat.label — and, on a card with no stat, the title — is drawn large. That is where the disclosure belongs.

Either fold it into the label ("data given · illustrative recreation") and keep the number as the hook, or drop the stat and make the disclosure the title outright. The second reads better at small sizes; the first keeps the hook. Decide with the target in mind, then check it by scaling a frame to 360p and looking at it — not by reading it at full resolution, where everything is legible.

Limits to state up front

One voice per playbook. tts.voice is global, so a customer voice plus a neutral closing voice is not possible yet. Write the whole narration in the customer's voice, or wait for per-segment voices.

No mid-roll cards yet. A pull-quote or a metric between segments needs segment.card, which does not exist. Use the opening and closing cards.

No split screen. Before and after run one after the other, never side by side. Sequence the contrast instead of compositing it.

Optional: generated B-roll

Generated clips (bytedance/seedance-1-5-pro through llm4agents) can open or close a testimonial with context a browser cannot show — a workspace, a counter, an abstract of the transformation. It is an accessory, not part of the format: the proof is the screen recording.

Two rules learned the hard way: never generate a screen with content on it — the model renders unreadable text, and once put a confirmation tick on the back of a phone — and never generate a person who speaks or testifies. Interfaces come from ndemo; people are context, not witnesses.

Checklist before rendering

  • Is provenance declared, and does the narration stay within what that route allows?
  • Is the narration first person, past tense for the pain, and specific?
  • Does the opening card lead with the result rather than the person?
  • Is the voice different from the product demo's?
  • Does the video show one concrete before/after, not a feature tour?
  • Is the disclosure legible at 360p, and present on the opening card when the route is illustrative?
  • Does the last frame commit to something — a result, a next step — rather than a generic CTA?

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.