Show HN: Transcript-critic, Claude Code skill: transcribe and critically analyze

A Claude Code Skill that transcribes audio files or

YouTube videos using whisper.cpp and then generates

a structured, critical analysis of the transcript -- all from a single slash command.

Run /transcribe followed by a .vtt file, an audio file, or a URL (including YouTube and other yt-dlp supported sites) and Claude will handle

the rest: downloading, converting, transcribing, and producing a detailed markdown summary.

Prerequisites | Installation | Usage | How it works | Transcript summary prompt | Personal Project Disclosure

This software was developed with the assistance of AI (Anthropic Claude). It is

provided "as is", without warranty of any kind, express or implied. Use at your

own risk. The author assumes no liability for any damages arising from its use.

Copy SKILL.md into Claude Code's global skills directory:

In transcribe.sh, update the following variables at the top of

the script to match your local whisper.cpp installation:

Note

In SKILL.md, the paths to transcribe.sh and ANALYSIS_PROMPT.md are referenced as instructions for Claude. If you clone this repository to a location other than ~/github.com/jftuga/transcript-critic/, update those paths in SKILL.md accordingly.

Once installed, use the /transcribe slash command inside Claude Code:

Claude will:

The skill is composed of two parts:

SKILL.md -- The skill definition file. It tells Claude how to handle each input type,

to invoke transcribe.sh, and how to produce the final analysis. Claude Code reads this

file when the /transcribe command is invoked.

transcribe.sh -- Accepts either a local audio/video file or a URL. Local, non-audio files are

converted to MP3 via ffmpeg; URLs are downloaded and extracted as MP3 via yt-dlp. In both

cases, whisper-cli is run to produce .txt and .vtt transcription files. Because yt-dlp

is used, audio can be downloaded and transcribed from

many websites -- not just

YouTube.

The file ANALYSIS_PROMPT.md is the prompt template that drives the analysis

step. It instructs Claude to produce a structured markdown document with the following sections:

This prompt is effective because it enforces objectivity because Claude is explicitly told to analyze

based solely on the transcript content without injecting prior knowledge about the speaker or

topic. It requires timestamp citations throughout, grounding every observation in a specific

moment. The evidentiary notes and logical fallacy sections push the analysis beyond simple

summarization into critical evaluation, categorizing claims by the strength of their supporting

evidence and surfacing reasoning errors. The result is an analysis that is structured,

verifiable, and avoids the uncritical paraphrasing that plagues most AI-generated summaries.

This program is my own original idea, conceived and developed entirely:

This project does not relate to my employer's actual or demonstrably

anticipated research, development, or business activities. No

confidential or proprietary information from any employer was used

in its creation.