ForgeKitFORGEKIT

Build Log

ForgeKit v2 — the loop made visible

June 26, 2026

Instead of building a better OS, we asked what the OS looks like when someone else holds it. The answer was a three-file Markdown instrument. Then we ran it on a real task — Flyback source research — and it worked on the first try.

ForgeKit OSLoop RunnerFlybackInstrument
3
instrument files built (README, template, example)
1
dogfood session completed (Flyback source research)
7
aerial imagery sources evaluated
2
next candidates identified (OGRIP, ODOT)

Timeline

Start
Ignite — v2 design challenge
Zeb: 'let's get crazy and build ForgeKit v2.' Ran ignition, Anvil Check on what v2 means.
Design
B vs C — transferable loop vs new engine
B already exists (forgekit-explorer). C is the build instinct. Chat introduced the C4 frame: Loop Runner.
Pivot
Scaffold challenged immediately
Next.js app scaffolded, then Zeb and Chat challenged: is this an app or an instrument? Markdown first, app second.
Build
Instrument contract + three files
Session Instrument Contract written. README, SESSION_TEMPLATE, EXAMPLE_SESSION built in loop-runner/instrument/.
Dogfood
First real session — Flyback research
Ran the instrument on Flyback source expansion research. 7 sources evaluated. Instrument held.

What shipped

loop-runner/instrument/README.md — what the instrument is, how to use it, the one rule (Evidence ≠ Judgment)

loop-runner/instrument/SESSION_TEMPLATE.md — Shape / Run / Close structure with ARTIFACT section as the exit gate

loop-runner/instrument/EXAMPLE_SESSION.md — worked example of the instrument running on its own creation

loop-runner/instrument/sessions/dogfood_session.md — first real dogfood session: Flyback source expansion research

Flyback source research artifact — 7 sources evaluated, OGRIP/OSIP as best near-term, USGS EarthExplorer as best long-term, ODOT as most interesting Ohio-specific

Next.js scaffold at loop-runner/ — parked, not deleted. Future shell when the instrument earns its UI.

The Markdown instrument tests whether the loop is useful at all. The app tests whether useful software can be built around it. Second question must come after the first.

Design principle that saved the session from premature appness