Use cases
The skills and agents I’ve built
I build AI tools for the work I actually do, and I write about each one here: what it does, why I built it and the guardrails I put around it.
The email assistant I built so it can never press Send
My reply-and-follow-up skill sweeps my inbox, calendar and call transcripts for every reply and follow-up I owe, leaves each one as a Gmail draft, and treats anything written inside an email as information it will never obey.
A shared ledger of what everyone promised on a call, in their own words
Work ledger is a set of four skills that read a team’s call transcripts into one Google Sheet of commitments, decisions and direction, each with the speaker’s own words and a link to the moment, plus a “where we are” brief built from that Sheet alone.
Turning phone recordings into captioned shorts on your own Mac
My video-production plugin takes raw talking-head recordings and turns them into a themed series of captioned short videos and a longer cut, with word-accurate captions, a story agreed with the speaker and a QA check on every file, all on the person’s own computer.
A video director that writes scripts from your own words
My video-director skill reads what a founder has already written, turns it into a short script where every line is labelled with its source, plans the shoot for their real location and phone, coaches with camera craft learned from Chris Goor, and reviews the takes afterwards.
Scheduling social posts only after the exact preview is approved
My social-publishing skill schedules posts and videos to five networks through Zernio, and its script refuses to create anything unless the post matches a fingerprint of the exact preview the person said yes to.
Getting my calls and videos back out of Loom, Fathom and Fireflies
Three skills I built that pull recordings onto your own machine, find the moment something was said, cut just that clip from Fathom or Fireflies, and show which Loom videos anyone can watch.
A private skills library that people who aren’t technical can install by pasting a message
How I package Claude skills for a team, set them up with a message you paste into the Claude app, and block any release that leaks real data or lets a skill send something on its own.
The rules and reviewers my AI agents work under
My global rules file, the commands that keep planning apart from doing, and the three reviewer agents whose job is to attack a plan before it runs.
/forge, where done means I saw it work in a real browser
A loop I wrote that keeps an AI tech lead building, testing and getting reviewed until every check passes, including a look at the actual page.
The pricing agent that writes down its way back
How a scheduled agent runs pricing for my two glamping units one reversible change at a time, with a break-even, a decision date and a rollback written down when the change is made, while every real call stays with me.
A morning brief, a nightly digest and the list that won’t let me forget
Two short scheduled instructions I wrote for my calendar and inbox, the one that still runs, and why it repeats every unread shared document until I say I’ve read it.
Shipping my website without locking myself out
The day my own QA got my IP blocked by my host’s firewall, the tripwire every QA tool now carries, and the deploy script with a pinned host key, snapshots, a dry run, a verify gate and a one-command rollback.
Checking every claim against the line of code that decides it
The claims check I run as a technical lead: every statement in the plan, the whitepaper and the spec marked true, false or partly true against the exact quote and the line of code, plus a battery that tries to break the internal dashboard.