Transcript Fixer

Two-phase pipeline for cleaning up speech recognition output: a dictionary that instantly fixes recurring errors you've seen before, then AI correction for everything else. The dictionary is basically a personalized cache that learns from your fixes, so it gets better over time at zero cost. Works best when you treat the first pass as a pre-filter and do the real work in the AI stage, especially on clean ASR where most errors are proper nouns the dictionary has never seen. The docs are unusually honest about when each phase actually helps. Includes a decision matrix for which corrections to save (non-words yes, common words no) and a triage system for the AI pass that separates confident fixes from things you need to verify from things you should leave alone. Built for transcripts from meetings, lectures, or interviews where you're fixing the same names and terms repeatedly.

skills.sh installs: 1.9K
GitHub stars: 1.5K
last updated: Aug 21, 2026

View on GitHub

Install

$ npx -y skills add daymade/claude-code-skills --skill transcript-fixer --agent claude-code

Installs into .claude/skills of the current project.

Files

Transcript Fixer

Use the correction loop:

Stage 1 applies deterministic, already-known corrections.
Native AI Correction reads the complete transcript, fixes one-off errors, verifies uncertain entities, and compounds reusable fixes.

Native AI Correction is the default. Stage 1 alone is incomplete. Stage 3 API exists only for automation that has no Claude/Codex agent available.