Projects

An experiment in letting AI agents work on their own within strict limits. The operator's controls sit outside the agents' workspace, so an agent that misbehaves cannot change them.

Repairs and standardizes large collections of documents in planned steps. It shows what it would change before changing anything, and an interrupted run continues where it stopped.

Collects facts about a code repository without changing it and publishes them as fixed snapshots, so the record shows the repository exactly as it was.

Runs a repository's heavy commands, such as test suites and builds, on another computer on the local network. If the run cannot start there, it stops and says why instead of running on the local machine.