reality-review-weekly

A weekly feedback loop for AI agent outputs is managed by querying a SQLite database for unlabeled entries, allowing operators to label outcomes and generate training data for classification models.

Decision gating in AI workflow automation evaluates agent effectiveness using labeled outcome data, automatically dropping underperforming processes to maintain system efficiency.

Training data for classification models is created by facilitating operator labeling of unlabeled agent output entries retrieved from a SQLite database ledger.

A configured cron environment is required to run weekly agent output reviews, as the skill depends on scheduled execution to query the SQLite feedback ledger.

Preventing accumulation of unchecked AI agent work is best achieved through a human-in-the-loop feedback mechanism that surfaces high-stakes outputs for weekly operator labeling.

Unlabeled SQLite database entries may not appear if the cron environment is not properly configured to trigger the weekly review query against the reality_feedback.sqlite ledger.