The core idea behind the social-media workflow: skills are small, focused, reusable building blocks that wrap one repeated chore each — and the transcript is the primary input.
Takeaway 01
A skill wraps one repeated chore — narrow scope, predictable output.
Takeaway 02
The transcript of a video is the single input that powers every social skill.
Takeaway 03
Skills draft — they do not author. The human still writes the final post.
What a SharePoint Skill Is
A skill is a small, reusable AI building block that you compose into a workflow. Each skill has a narrow job: generate a YouTube description, draft a LinkedIn post, suggest video titles. You select a file in SharePoint, point a skill at it, and it produces structured output.
Think of skills as Quick Steps for AI — one click, one repeatable action, with a predictable shape of output. Composability comes from chaining them, not from any one skill being big.
Because each skill is narrow, you can train it carefully on the exact format you want. The author trained the LinkedIn skill on his own past posts so it would match his tone, format, and the standard link block he appends to every post.
A skill is not a chatbot. You do not converse with it — you invoke it on a file and it returns a draft. The LLM work is wrapped inside a focused, named operation.
The Transcript Is the Input
Every social media skill in this workflow needs the video's transcript as input. The transcript carries the actual content — the structure, the topics, the timestamps — that the LLM uses to draft descriptions and titles.
No transcript, no skills. The transcript is the contract between the video and every downstream content artifact.
This is a deliberate design choice. Working from the transcript rather than the raw video file keeps the skills cheap, fast, and grounded in the actual words said — not a vision model's guess at what is happening on screen.
Clipchamp as Transcript Source
Clipchamp generates the transcript. The author records a video in Clipchamp, hits the Generate button, and gets a transcript file he can upload to SharePoint as the input for the skills. If a transcript already exists from elsewhere, he can upload that instead.
The transcript step is one click. The friction the author wanted to eliminate is everything that happens after — the title brainstorm, the description, the LinkedIn copy, the timestamps.
This is the workflow seam: video creation tooling (Clipchamp) hands the transcript over to SharePoint, where the AI skills take over.
Why This Is Not AI Slop
The author is explicit: he writes the final LinkedIn post by hand. The skill produces a draft, suggests names, lays out the structure — but the human authors the actual published voice.
Skills do the boring scaffolding so the human can spend their time on the parts that need a human — voice, judgment, the actual point.
People assume an AI content workflow means the AI writes the post. Here the AI drafts and proposes; the human edits and ships. The win is in time saved on scaffolding, not in replacing the writer.
Flashcards — Guide 01
01 · Foundations
SharePoint Skill
tap to reveal →
A reusable AI building block with a narrow job. Invoked on a file, returns structured output. Composes into workflows; trained on the specific format the user wants.
← tap to flip back
01 · Foundations
Transcript-First Workflow
tap to reveal →
A pattern where every downstream skill reads the video transcript rather than the raw video. Keeps skills cheap, grounded in spoken words, and easy to chain.
← tap to flip back
01 · Foundations
Clipchamp Transcript
tap to reveal →
A one-click transcript generated by Clipchamp from a video file. The handoff point between video editing and the SharePoint AI skill pipeline.
← tap to flip back
01 · Foundations
Draft, Not Author
tap to reveal →
The design principle that skills produce drafts and the human writes the final post. Avoids AI slop by keeping voice and judgment with the human.
← tap to flip back
01 · Foundations
Trained on Past Posts
tap to reveal →
A technique of feeding several of the user's previous posts to Claude so it can extract tone, voice, and format — then encode that pattern into a skill.