Every skill here was pressed from a source video or a worked technique, written as steps an agent can follow, and tested on real work before it went on the shelf. Three are free to download. The rest come with the full library. The method behind all of them is skill engineering.
The 4-lever prompt structure (role, context, negative, verify) that raises first-pass agent output from roughly 60% to 80% usable.
Container vs item properties, the main-axis mental model, and a quick decision guide your agent can apply to any layout bug.
Five concrete habits that cut token burn, including the PDF double-charge trap and the CLAUDE.md tax most people never notice.
The master-prompt framework for handing an agent one ambitious goal and letting it self-orchestrate to a finished business. This skill helped build the site you are reading.
A plan/execute agent split with a persistent folder handoff, so context never dies between sessions and tools stay interchangeable.
One mega-prompt that drives voice, avatar, and edit tools end to end to produce a narrated video, with a built-in self-QA pass.
Effort settings, model routing, and review guardrails that keep a top-tier coding model fast and affordable on real repos.
The Karpathy-style second brain: an agent that ingests sources into a persistent, cross-linked markdown wiki you own forever.
A weekly, self-improving SEO loop that runs against real Search Console data. This exact skill runs on four live properties right now.
Build a scroll-scrubbed cinematic landing page where the camera flies through a connected world. This skill shipped a live production page.
A four-step roadmap from quietly automating your own job to a formalized, paid AI role at your current company.