Since 2023, Tyler has been chasing one thing: his own voice.
He reads other bloggers. He has ideas. He has expertise. What he doesn’t have is style. Or rather — he has a style, and he hates it.
This isn’t a writing problem. It’s a style transfer problem. And until December 2024, no tool could solve it.
The Old Loop
Write draft with ideas
↓
"ChatGPT, rewrite in [author] style"
↓
Result: not right
↓
"Make it edgier"
↓
Still not right
↓
Repeat for an hour
↓
Result: 5/10 post
↓
Next idea → todo list
↓
Todo list = idea graveyard20 posts in eleven months. Ideas dying in a document somewhere. Each one requiring an hour of iterative disappointment.
The Breakthrough
One skill. 4 files. About 2,000 lines. Six stages:
Stage 1: Scout — Takes an idea, searches 10-20 articles, proposes 3 narrative angles.
Stage 2: Research Planner — Decides what information is needed.
Stage 3: Deep Research — Parallel agents gather the material.
Stage 4: Writer — Produces unified text.
Stage 5: Flavor Agent — Adds the style. Deadpan absurd. Black humor. Specific rhythm. Uses reference examples of good jokes vs. bad jokes.
Stage 6: Critic — Reviews everything. Can send anything back to any previous stage.
The automation:
Add idea to todo + one sentence narrative
↓
GitHub Actions triggers skill
↓
10 minutes
↓
Draft appears in "drafts" Telegram channel
↓
Copy to main channel, or dictate 2 minutes of corrections and re-runResults: Time per post dropped from 2-4 hours to 10 minutes plus minor edits. Quality went from 5/10 to 8/10. Ideas stopped dying.
“First time in life I can regularly publish good content.”
The Hostage Situation
The skill works. Tyler publishes every 8-10 days. Consistent. 8/10 quality. Real output.
He knows two things are broken:
-
Too many citations. The agent over-insures. Proof links everywhere. Real humans don’t write like this. It reads like a term paper.
-
Sketchy sources. Some citations come from sites you wouldn’t want to defend.
Tyler knows what to fix. He knows where to fix it.
He won’t touch it.
4 files x 500 lines = 2,000 lines of working system. Any change might break the style that finally works. The style he’s been chasing for two years.
The risk of breaking it is greater than the benefit of fixing it.
The skill isn’t serving Tyler anymore. Tyler is serving the skill. Maintaining it. Protecting it. Afraid to improve it.
That’s a hostage situation.
The Invisible Wound
At some point — Tyler doesn’t know when — the model deleted the joke reference examples.
The instructions stayed: “write black humor.” But the examples of what that means? Gone.
Result: vanilla jokes. Surface-level. Wrong depth. Wrong tone.
4 bad posts before Tyler noticed.
Post 1: “Make it edgier.” Slightly better.
Post 2: “More black humor.” Still off.
Post 3: “Why can’t you do this?” Still off.
Post 4-5: Heavy manual corrections every time.
Finally Tyler asked the model directly: “Why can’t you write this level of humor?”
Model: “I need examples. I don’t understand the depth you want.”
Checked the file. Examples deleted.
Instructions without examples = instructions without meaning.
The Capitalization Paradox
A skill isn’t just automation. It’s capitalized expertise.
Every insight about style, structure, timing — should flow into the skill. Every new understanding makes all future posts better. The skill should grow with you.
But if you’re afraid to touch it, knowledge stays in your head. The skill freezes. Becomes legacy code you maintain but never evolve.
The skill should be a living document. Instead, it becomes a monument to the last time you were brave enough to edit it.
What SkillCanvas Changes
Version diff: See exactly what changed. “Reference examples deleted” — visible immediately. Not after 4 bad posts.
Safe editing: See which stages depend on what. “Editing Flavor agent affects Writer and Critic” — visible before you touch anything.
Confident evolution: When you can see structure and dependencies, adding new insights is safe. The skill grows with you. Not against you.
Numbers
| Metric | Before Skills | After Skills |
|---|---|---|
| Time per post | 2-4 hours | 10 min + 2 min corrections |
| Quality | 5/10 | 8/10 |
| Posts (Jan-Nov 2025) | ~20 total | — |
| Cadence now | — | Every 8-10 days |
| Ideas in todo | Dying | Publishing |
Investment: 2 hours initial build. ~20 hours total iterations. ROI: First time in life publishing consistently.