Taste skill / content humanizer
First met in aicp-c3-basecamp-04 · Productivity & source apps
Taste skill / content humanizerFirst met in aicp-c3-basecamp-04 · Productivity & source apps
Instruction packages that carry aesthetic and voice judgement — the difference between correct output and output you'd publish.
Generated text or design is technically fine but reads as machine-made and you don't want to write the style rules yourself.
The humanizer is based on Claude's own official humanizer instructions
Encodes taste as reusable text, so the same judgement applies across tools
Rewrites generated content to read less like a model wrote it
Supplies design and voice judgement as a system prompt to an agent
Doesn't make bad content good — it fixes register, not substance
The last pass before anything generated goes public — the newsletter step in BC4's automation.
Productivity & source apps
The KB's voice is set by the extraction protocol and Paul's design review, not by a borrowed skill.
The idea is more useful than the artifact: a short written statement of the KB's voice would do the same job the humanizer skill does for the newsletter.
current
skills-sh: distributed there
n8n: used as the agent node's system prompt in the newsletter workflow
| Session | Coverage | Moment | What happened |
|---|---|---|---|
| Session 3: Building Elegant Landing Pages | demonstrated | 1:06:49 | Drove the interior-design premium build from one prompt + link; generated its own images; BC4's bake-off skill returning as a full build driver |