What problem does it solve? Lecture transcripts produced by speech recognition contain misrecognized legal terms, article numbers, and case names, and manually tracking which transcript parts have been corrected across many folders is error-prone. This Skill enforces strict evidence-based correction rules and automates the bookkeeping of correction status. ## Core Features & Use Cases - Evidence-Based Correction: Corrects only clear typos and terminology errors while preserving original wording, requiring a two-line source citation for every change and marking unclear segments as [불명확] instead of deleting them. - Correction State Tracking: Records reviewed versus generated corrections per transcript in transcription_log.json, detects duplicate transcripts by SHA256, and repairs stale paths. - Batch Automation: Lists uncorrected transcripts, generates heuristic corrected drafts from a replacement dictionary, and batch-marks completed corrections with progress and index hooks. - Use Case: After transcribing a civil law lecture into part files, run list_uncorrected.py to see pending parts, correct them against the textbook, then run mark_corrected.py to log completion and update study progress. ## Quick Start Ask the assistant to correct a specific transcript part file against the textbook source and then mark it as corrected using the provided script.
