Gomny: A Publishing Pipeline Built on Receipts

The Problem

Most publishing pipelines, including the company scale platform I built, solve the writing problem with a capable agent system behind a company’s infrastructure. gomny asks a narrower question: what does the smallest version of that look like for one person, with no team, no dashboard, and a human explicitly in the loop on every single post?

The Approach

gomny is deliberately the opposite of an agent factory in scope: one CLI, six pipeline stages (capture, extract, enrich, draft, gate, publish), and a hard rule that nothing reaches GitHub without an explicit human approval typed at the terminal. Every insight the drafting stage uses has to appear verbatim in the source session file, or it gets dropped. Receipts take priority over generated claims. Where an agent factory optimizes for scale across clients and asset types, gomny optimizes for trust in a single, small, auditable loop.

The Outcome

This site’s own articles run through gomny end to end, including this case study index and the rest of the site structure around it. It is also part of a pattern I have now built twice: once at company scale inside an agent factory, and once alone, at the opposite end of the scope spectrum, deliberately minimal instead of deliberately powerful.