leak-learner

Semi-automating owner feedback learning involves coordinating correction logging, candidate generation, and gated persistence to global rules. This loop captures corrections and applies updates to quality-rules.yaml and lessons only after explicit owner approval.

Session-scoped learning prevents persistent mislearning by enforcing non-persistent memory guardrails. It isolates feedback processing within the current session and requires explicit approvals before applying any updates to global rules or lessons.

The best way to manage learning candidates is integrating them into a nightly morning-review. This workflow presents generated correction candidates for owner decision in daily reports, driving governance through a structured approval gate.

Explicit owner approval is required to update quality-rules.yaml and lessons. The semi-automatic learning loop uses an approval gate to separate global rules, skill lessons, and client lessons, ensuring human governance before persistence.

You can separate global rules, skill lessons, and client lessons using an approval gate. This structure ensures that feedback corrections generate candidates specific to each category before any changes are gated for persistence.

Session memory does not persist corrections automatically due to non-persistent memory guardrails. These constraints prevent unreviewed updates, requiring a gated process to log corrections and generate candidates for explicit owner approval.