Gate decisions that teach the next run
Every request-changes loop is expensive knowledge. agI3R keeps it: decisions made at gates are captured as records, projected into a knowledge base, and fed back into the prompts of future agent runs.
How it works
Approvals, rejections, and request-changes feedback are already events — the knowledge base is built from what humans actually decided, not from a survey afterwards.
The decisions KB is a projection of the same immutable event stream as the boards — replayable, idempotent, and never a second source of truth.
Relevant prior decisions ride along in future agent prompts — the agent that drafts your next spec or diff already knows what your reviewers rejected last time.
Nothing learned in one tenant informs another — cross-tenant learning is off by default and stays off pending explicit consent and compliance review.
Human control
The KB only ever contains what humans already decided at gates — consuming it never adds authority or bypasses a gate.