This site
Personal · 2026
A portfolio built by one person directing Claude Code agents, with the plan, the rules and the gates left on show.
Proves: The harness, applied to itself
One person directing Claude Code: contracts first, then agents in parallel, then gates.
This is the thesis, applied to the site you're reading. I wrote the brief and made the decisions. A lead Claude Code session settled the contracts: a conventions file, a frozen content schema and a fact ledger. Agents built everything else, each in a directory it owned.
The lead reviewed and committed every agent's work, one commit per agent.
Read from the repository when this page was built.
Scaffold, then design tokens. An identity checkpoint, where I chose one of three candidate skins from screenshots. Then the frozen content schema, the fact ledger and the conventions file.
Routes and whole-page stories.
One review each for voice and facts, for accessibility and responsive design, and for code. Then the fixes and the final gates, by the lead.
Every exported component has stories beside it. A missing story fails the check.
axe runs on every story and every route.
Stories and pages are screenshotted and compared against a stored baseline.
Every number, date, title and employer in the copy must match a ledger entry with a source. A figure with no source fails the test.
A test checks the contrast of the color tokens in both themes.
Strict TypeScript and Biome, clean before any agent reports done.
What stayed with me: the brief, the decisions (scope, which projects, the look, what could be said about Reddit), the choice of skin, and approval.
The design system is documented in a public Storybook. When the model improves, none of the above changes and the output gets better. That is the argument for building the harness.
Personal · 2026
A portfolio built by one person directing Claude Code agents, with the plan, the rules and the gates left on show.
Proves: The harness, applied to itself