A harness, not an opinion
satelle is a harness. It runs a repo's process as configuration. The binary holds only the mechanism. The process itself, its gates and its opinions are written by the people who own the repo, and they live in the repo under .satelle/.
Mechanism and configuration
The mechanism is what the binary does: it runs agents and enforces gates. Everything else is configuration that a team writes: workflows, skills and the rubrics its reviewers judge against.
A few rules keep that line clean. No gate is written as code. The route a change takes is configured, not decided case by case. Quality comes before speed. A user's configuration is never compiled into the binary. Only the structure satelle needs in order to work is fixed. Everything else can be changed.
Your process, not ours
satelle is built for other people's repos. A developer or a company writes its own workflows, skills and reviewer rubrics for each repo, and satelle enforces them. It does not ship one process that everyone has to follow. The satelle repo itself is built with satelle, under its own configured process.
Any agent
satelle is not tied to one agent either. It drives Claude, Grok, pi and Cursor, over whichever transport each uses. Every mechanism works for each agent, or satelle says plainly that it is not available for that one.
In practice
About 20 repos on one machine run satelle today. They share one machine-wide catalogue of agents, and each keeps its own workflows.
That is what the earlier posts in this series relied on. Changing a reviewer's model was a change to that catalogue, not a change to satelle.
To be continued.