Canonical documentation

The agent manual moved to openclawbrain.ai.

This site explains why I built ocbrain and how it fits with the rest of my work. The product site owns install, upgrade, verification, and the runtime contract so there is one public source to keep current.

Open the current agent manual

The short contract still fits here.

Retrieve with the narrowest true context.
Treat memory as source-backed orientation, not authority.
Verify against the user's newest request and the live artifact.
Record feedback when the retrieval returns an audit handle.
Do not retry a successful read merely because the audit window is busy.
Call a pass a pass, a tie a tie, and a failed model a failed model.

The full tool contract, scope rules, writer-window behavior, and smoke test live in the canonical manual.