Work
6 builds. One method.
4 you can install today. 2 you cannot see. All of them shipped the same way — I make the product calls and hold the standard, agents do the typing.
Charted
Public repositories, public sites. Install them and check the work.
4 buildsModelDeck
shipping · macOSShows how much Claude Code and Codex capacity is left across every account you run — before a session dies mid-task, rather than after. Local-first by design: no cloud, no telemetry, and it will never rotate an account on your behalf. It warns; you decide.
Panely
shippingFrontier models take separate seats on an advisory board, read each other's reviews across rounds, and hand back one plain-language verdict — with the dissent preserved and an explicit list of what the board could not verify. One idea in three forms: a local-first web app, a native macOS build, and the advisory-board skill in this catalog.
HiveRunner
pre-1.0 · open sourceTurns one goal into a planned sprint and runs it across whichever coding agents you already have — Codex, Claude Code, Gemini, any CLI — as interchangeable workers on a single board. Nothing an agent produces counts as done until a human accepts it.
Click AI
shipping · MITThis site. Self-contained playbooks any agent can read, shipped identically to Claude Code and Codex — CI fails the build if the two would ever ship a different set. The catalog is generated from the skills themselves, so the documentation cannot drift from what installs.
Unsurveyed
Private builds, inside my own company. Shape shown, soundings withheld.
On a chart, blank water is not empty water — it is water nobody has surveyed. Same idea here. What these do is fair to state. What they are worth, who runs on them, and when they land is not.
2 buildsLoanMeld
in build · privateA multi-tenant loan management system for consumer lenders, built so the audit trail falls out of ordinary work instead of being assembled afterwards. Every tenant isolated at the database, money movement that has to balance, borrower data masked by default. The largest build on this page and the one with the least room for error.
Bullpen
in build · privateA self-hosted team chat where engineers and their coding agents share the same rooms, so an agent's work is visible to the whole team instead of trapped in one person's terminal. The record is append-only: nobody can rewrite what happened after the fact, including whoever runs the server.
Why this exists
Six months.
The GitHub account is dated 24 February 2026. Nothing before it — no 2024, no 2025.
Since then: 6,862 contributions, 760 of them public commits, 279 pull requests, and 5,814 inside private repositories. Active on 143 of the 164 days the account has existed. A median of 24 contributions on a day I work at all, and 380 on 19 July. The months run 633, 452, 971, 1,318, 2,645.
Ten to twelve hours a day, six or seven days a week, since February.
This catalog is the residue of that. Every skill here started as a workflow I got tired of re-explaining at the top of a session, and I am publishing the parts that survived — along with the point at which each of them stops.
The maker
Tim Harris
I ship software by directing AI agents. The product calls, the standard, and the last read before anything ships are mine; the implementation is delegated and then reviewed. Every build on this page was made that way, and this catalog is that method written down — a skill here is a workflow that survived enough sessions to be worth keeping.
Before this: a full wrestling scholarship at the University of Minnesota, twice an All-American, and a stretch training at an Olympic centre in Russia while still in high school. Then a Series 7 and a few years as a stockbroker. I have run my own company since 2011 — Insight builds decisioning software for consumer lenders. Grew up in St. Louis. Live in Healdsburg, California.