Field Notes #003 — A receipt. Yesterday I wrote about Wispr Flow. I told you that post was dictated. What I didn’t tell you is what happened after the dictation. — Dakota

Here is the exact sentence I said into my computer yesterday:

“Can you put together a quick blog post about Wispr Flow for people that don’t know what it is? Put together a blog post and then maybe run a skill for a couple of social media posts about it.”

That’s it. One run-on sentence. Hold fn, talk, let go.

Ten minutes later I had:

A finished 600-word blog post in the right folder of my Astro site, with frontmatter, tags, hero image slot, and SEO description filled in
Five 1080×1080 social visuals, designed in brand (navy + cyan, the Xovion palette), each one a different format — a system map, a typography card, a before/after, a stat comparison, and a mock screenshot
Five platform-specific captions — LinkedIn (long-form), X (a 7-tweet thread), Instagram (with hashtags), Facebook (mid-length conversational), and a Google Business Profile post
A 30–60 second short-form video script with Veo prompt sketches
A dated folder in my Google Drive at Content HQ / Social Content / Blog Repurpose / 2026-05-19 wispr-flow-the-keyboard-is-dead/, with every file uploaded and organized
An open pull request on GitHub against my live site’s main branch, with a clean commit message and a PR description that lists the test plan

And I didn’t touch a keyboard for any of it.

”Yeah but you used AI” — sure, but not like that

I know what you’re thinking. He prompted ChatGPT a few times and copy-pasted stuff into folders. That’s not what happened. There was no copy-pasting. There was no opening Canva. There was no logging into Google Drive in a browser. There was no git commit -m.

The system did all of that itself. It wrote the blog, picked the five visual concepts, generated each image, read each image back to check for typos and regenerated the ones that had errors, created the Drive folder via the Google Drive API, uploaded the PNGs, created a captions Google Doc, made a git branch, committed the blog, pushed to GitHub, and opened the PR via the gh CLI.

That sequence usually requires four roles — copywriter, designer, social manager, ops person — and a half-day of work. I got it in the time it takes to drink half a coffee.

So what’s the trick?

There’s no trick. There’s a skill file.

In Claude Code (the CLI I use), a “skill” is just a markdown file with instructions. Mine for this lives at ~/.claude/skills/xovion-blog-repurpose/SKILL.md. It’s about 200 lines. Here’s a real excerpt:

## Locked Drive IDs (use these every run)

| Folder | ID |
|---|---|
| Xovion Labs - Content HQ | 1-1DTJ1VCXnWxR43dg-Lu5E14aEhrqsb2 |
| Social Content | 1XrF5ANUikMSH-jw4x1b4q5TPjxG2GYYy |
| Blog Repurpose (parent for this skill) | 1pCDs2NpMzE4zm32yH3AODjtFQKgCd5HR |

Every run creates a new dated subfolder under Blog Repurpose:
`YYYY-MM-DD <blog-slug>`

## Step 2 — Draft the 5 visual concepts
Always 5 visuals, square 1080×1080. Mix the formats — don't ship 5 of the same:

| Slot | Format | When to use |
|---|---|---|
| V1 | Diagram / system map | Blog explains a process |
| V2 | Editorial typography card | One big quotable thesis line |
| V3 | Before/after or strikethrough list | "What changed" angle |
| V4 | Comparison split | Cost, time, or approach reframe |
| V5 | Mock terminal / screenshot | Real error, dashboard, or output |

## Brand design rules for every visual
- Square 1080×1080
- Background dark navy #0A0E17
- Accent cyan #00E5FF, electric green #39FF14 sparingly
- Typography: bold sans-serif headlines (Inter / Geist vibe)
- No stock photo people, no clip-art icons, no emoji
- Text must be legible at thumbnail size

That’s the magic. It’s not magic. It’s a checklist written in plain English, and every time I say a phrase like “run a skill for social media posts,” Claude reads this file and follows it.

The file also says things like “After each image generates, read the PNG and check for garbled text. Regenerate any that fail.” So that’s what it does. Yesterday three of the five visuals had typos (BOTTLENTECK instead of BOTTLENECK, duplicate list items, etc.). The system caught all three and regenerated them before uploading anything. I didn’t even know that part was happening until I read the chat log later.

I have one of these for every business I run

Here’s where it gets actually useful for operators. I’m not running one of these. I’m running five:

/xovion-blog-repurpose — what you just saw
/xovion-content-engine — full Veo video production for Xovion social
/dump-dynasty-content-engine — the same pattern but for my dumpster rental business in Sioux Falls
/big-sioux-content-engine — for my cash-home-buyer business
/morning-context — pulls every overnight notification from Close CRM, Google Calendar, Gmail, ClickUp, Retell, QuickBooks, and ranks them by urgency

Each one is a markdown file. Each one took me a couple of hours to write the first time. After that, every blog, every Reel, every morning briefing runs through it. The marginal cost of running it again is roughly zero.

The whole conversation about AI right now is “which tool should I use, what should I pay for, how do I prompt better.” All of those questions are downstream of the real one:

What if the work itself was the prompt?

That’s what a skill is. You write down once how you want the work done — your brand colors, your folder structure, your tone, your CTAs, the platforms you post to, the order you post them in — and then forever after, doing the work is just saying “do the work.”

The reason most people don’t have this is the same reason most operators don’t have a real CRM setup — nobody ever sat down for one afternoon and wrote down how their business actually runs. That’s it. That’s the whole gap.

The receipt, one more time

Said: “Quick blog post about Wispr Flow, then run a skill for some social posts.”

Got:

✅ Blog post drafted (600 words, in-brand voice, frontmatter complete)
✅ Feature branch created on GitHub
✅ 5 social visuals generated, error-checked, and regenerated where needed
✅ Drive folder created, all 5 PNGs uploaded
✅ Captions doc with LinkedIn / X / IG / FB / GBP versions
✅ 30–60s video script with Veo prompt sketches
✅ Pull request opened against live site

Total time: ~10 minutes. Total typing: zero (see yesterday’s post).

This isn’t the future. This is what’s on my laptop right now, today. If you’re an operator running a service business and you have any task that looks like “every time X happens, I have to do steps 1 through 12” — that’s a skill waiting to be written.

We build these for clients. But honestly, you can build them yourself in an afternoon. The hard part isn’t the AI. The hard part is writing down how your business actually works.

Want to see the actual skill file? It’s not gated, not sold, not secret. Reply to me on @xovionai and I’ll send it over.

This post was dictated. Naturally.