transcription-corrector

Provide the raw transcript and a YAML dictionary of correct term spellings. The Skill reads the full text, replaces only high-confidence mis-transcriptions confirmed by context and form-drift evidence, and writes a corrected Markdown file plus a correction log.

The dictionary lists target spellings like WorkBuddy or Claude Code, acting as a confidence weight rather than a trigger. A replacement happens only when the dictionary entry, clear contextual pointing, and form-drift evidence all hold; otherwise the original text is kept.

Yes. TXT inputs are converted to Markdown first: speaker-plus-timestamp lines become bold headings and paragraph spacing is normalized, while all original wording, blank lines, and decorative characters are preserved unchanged.

Yes, filler-word removal runs by default using a whitelist of pure fillers. Attitude markers, logical connectives, segment-final rhythm markers, and fillers adjacent to key nouns are preserved to avoid changing speaker intent.

Do not use it for rewriting transcripts into course chapters, reports, or summaries from scratch, or for creating content from blank material. It only proofreads and lightly polishes existing transcripts without changing facts or structure.

No. The original file is never touched. Corrected output is written to a timestamped archive directory and mirrored as a _corrected.md copy next to the source file, with timestamp suffixes used to avoid overwriting prior versions.