Everything here is something I use in production, extracted and published. Working instructions and code, not tutorials about them. Newest first.
The rule I hold myself to: nothing goes on this page unless it already runs. If a resource stops being true, it comes down rather than getting quietly patched. Where a tool has a limit, the limit is on the row, not behind the click.
I capped how large a Claude Code session can grow before it compacts, then measured what happened across 1,229 sessions. The write-up has the numbers and the one line to copy into your own settings.
The cost before-and-after can't be measured any more, because Claude Code deleted the older transcripts. The article says so and makes no cost claim.
The dictation app I use every day, with a meeting recorder built into it. It captures your microphone and the other side of the call as two separate tracks, then hands you a speaker-separated transcript. Everything runs on your own machine.
Source only, and it needs macOS 14.2 or later on Apple silicon. There is no download, because the build is signed with a certificate only I have. You change one line and build it yourself.
The working instructions for driving a Webflow site with an AI agent. CMS architecture, bulk updates, safe publishing, accessibility and SEO audits, custom code, the CLI and the MCP setup underneath it.
These are the skills my own agents run against live Webflow sites. Nothing here was written for the giveaway.
An MCP endpoint that puts live Meta ad data where you already work. Connect it once, then ask Claude or ChatGPT what a brand is running and get the answer in the conversation instead of going to look for it.
Built for Wilow, one of my own businesses, and free to connect.
Type a brand, see every ad it is running right now. Filters, the video variants sitting behind a single concept, and a page for each brand and each vertical. It runs in the browser with no account and no card.
Search is open to everyone. The deeper strategy breakdown asks for an email, and giving one lifts the daily search cap. That is the only gate on this page.
More coming: the drift check that proves an agent fleet still does what its documentation claims, and the autonomy ladder that decides what an agent may do without asking.
Taking these and running with them is genuinely the point. If you would rather not, that is the work I do.
Get in touch