What problem does it solve? Code reviewers rarely get feedback on whether their comments land well. This Skill turns external PR reviews into deliberate practice by recording a prediction for each comment (accept, pushback, or ignore) and scoring those predictions after the PR closes, revealing calibration patterns over time. ## Core Features & Use Cases - Prediction Logging: Annotate each posted review comment with a predicted outcome, confidence level, and rationale, stored as JSONL telemetry. - Automated Scoring: A script classifies actual outcomes via the GitHub API by inspecting review threads for author replies and resolution status. - Calibration Reports: Generate accuracy, per-outcome precision/recall, Brier score, and confidence-bucket statistics to spot anti-patterns like high ignore rates or overconfidence. - Use Case: After reviewing a colleague's PR on an external repository, log predictions for each comment, then run the scoring script a week later to see which comments were accepted, debated, or ignored. ## Quick Start After I post review comments on an external PR, help me record a prediction with confidence for each comment and log it to the telemetry file.
