Harness — a personal agent platform

1. Thesis

FridaiOS — my day-to-day operating harness — is a personal agent platform with the same shape as an organisational agent deployment: one knowledge source of truth, skills as portable verbs, staged workflows with human gates, and observability and cost governance running underneath. I built it to run my own work, which means every piece of it has been exercised against real, messy tasks rather than a demo dataset.

2. Components

3. Claude Code maturity

Claude Code has been my daily driver since early 2026. Between April and July 2026 I put ~430 commits across 9 self-directed repos through it — skills, hooks, and MCP-client orchestration running in production personal use, not a one-off proof of concept. The platform pages on this site (this one included) were themselves built through that harness.

4. Equivalence

A fair question is whether a skills-and-hooks harness built on Claude Code is doing fundamentally different work from a graph-based orchestration framework like LangGraph, or whether the two are equivalent architectures wearing different clothes. I've written up that comparison in detail: the LangGraph equivalence brief.

For the underlying engineering — the router, the workflow dispatcher — see the exhibits directly: