—
Home
Probabilistic input. Deterministic acceptance.
Tsqoba is a brownfield agent runtime for existing repositories. On the paths it already owns, a change stays provisional until the runtime accepts it. That lets the system stop some bad writes early, run heavier checks later where they belong, and keep rejects, replays, and exceptions inspectable.
Current phase: core refactor + internal dogfooding. Public pilots are not open.
An agent multiplies what already exists
Most software work happens in brownfield systems. In that environment, a model learns from accepted code in the repo: interfaces, shortcuts, naming patterns, test habits, and tolerated failure modes.
That is why the acceptance path matters more than prompt polish. If small shortcuts keep landing, they compound. If cheap mistakes are rejected early, heavier checks happen later where they belong, and exceptions stay explicit, the runtime teaches a different next move.
Tsqoba is built around that practical problem. It does not try to make brownfield drift disappear. It makes more of the acceptance path inspectable and less of it implicit.
- accepted code teaches more than prompts
- small tolerated shortcuts compound quickly
- inspectable rejects are cheaper than silent drift
Where to read next
Read Articles for current mechanics, product boundaries, and trade-offs, including how owned write paths differ from CI and local hooks. Read Cases for one operating episode at a time: trigger, system response, outcome. Read Artifacts for the retained material behind those pages. Read Status for the current surface, recent changes, and the rough edges that still remain.