A forge that runs its own code

A forge you sent runs this forge's code until you switch it. Once switched, it starts from its own <org>/forge repository's main, so changes made there are what it runs. Its limits, sandbox and egress rules are set by this forge, not by its code, and they are the same either way.

Switch it

on the forge's row. Press Use this forge's code to switch back.

pick the forge. The picker says whose code each forge runs and when it last rolled back.

A running forge is put to sleep. Its next visitor wakes it on the new code.

Only a forge in its own cell, with a forge repository, can be switched. A forge that shares this forge's process is told so and stays as it is.

What a start does

  1. It copies its main into this forge's data, under own-code/<name>/<commit>.

The repository and the copy must each stay under 256 MB and 50,000 files. The copy must hold no symbolic links, a src/main.ts that parses and a forge.json.

  1. It starts the cell on that copy. If the copy fails its check, or the cell

does not answer in time, it tries the last commit that came up, then this forge's code.

  1. Each fallback is a Rolled back row on /forges and a

forge.rolled_back event, in fixed words. A commit that failed is not tried again until main moves.

How a child running its own code is contained: docs/THREATS.md.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs