Convert one or more local videos into clean, timestamped TXT transcripts without modifying the original media.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install video-transcript-txt
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install video-transcript-txt using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Video Transcript TXT?
Video Transcript TXT is an Openclaw Skills workflow for transcribing complete local video files into readable plain-text documents with start and end timestamps for every valid speech segment. It uses uv, ffmpeg, ffprobe, and openai-whisper, while defaulting to Chinese (zh) when no language is specified.
The skill is designed for reliable single-file and batch transcription. It preserves source videos, rejects incomplete downloads, stores Whisper intermediate files only in temporary system directories, prevents unauthorized overwrites, and reports transcript paths, segment counts, media duration, and actual processing time.
Video Transcript TXT Use Cases
- Transcribe a locally stored lecture, interview, meeting, webinar, or presentation.
- Extract searchable spoken content from one or more video files.
- Generate timestamped TXT files for editing, research, archiving, or knowledge management.
- Batch-process multiple videos with bounded parallelism and per-file status reporting.
- Measure both total successful video duration and real elapsed processing time.
- Re-run transcription safely without overwriting existing results unless the user explicitly authorizes
--force. - Process multilingual content by selecting the appropriate Whisper language code.
How Video Transcript TXT Works
- Validate each absolute input path, confirm it is a regular file, reject incomplete-download suffixes such as
.downloading,.part, and.partial, and verify that the output is a.txtfile. - Check that
uv,ffmpeg, andffprobeare available before processing begins. - Determine the default output path as
<video filename>_整理版.txtin the source video's directory, unless a custom output path is supplied. - Refuse to overwrite an existing final TXT unless the user has explicitly approved replacement and
--forceis provided. - Use
ffprobeto obtain the source video's media duration. - Create a system temporary directory for all Whisper intermediate outputs.
- Run Whisper through
uv run --with openai-whisper whisper, using the selected model and language, with generated JSON and other intermediate formats kept out of the video directory. - Read the JSON
segmentsarray, retain valid segments, sort them chronologically, and format each line as[HH:MM:SS.mmm - HH:MM:SS.mmm] text. - Atomically replace or create the final TXT, then remove the temporary directory on both success and failure while leaving the source video untouched.
- Emit the final TXT path, valid segment count, video duration, processing duration, and a machine-readable
SUMMARY_JSON:line. - For multiple videos, the main agent preflights all inputs, dispatches one video per sub-agent with a default concurrency limit of 2 and a maximum of 3, continues independent jobs after an individual failure, and summarizes successful and failed items.
Video Transcript TXT Setup
Requirements
Install or make available:
uvffmpegffprobe- The skill script at
~/.claude/skills/video-transcript-txt/scripts/video_transcript_txt.py - Network access or a populated Whisper model cache when the selected model must be downloaded
Verify the command-line dependencies before running:
command -v uv
command -v ffmpeg
command -v ffprobe
Run a single video
uv run ~/.claude/skills/video-transcript-txt/scripts/video_transcript_txt.py \
"/absolute/path/to/video.mp4"
The default output is created beside the source video as <video filename>_整理版.txt.
Choose an output path
uv run ~/.claude/skills/video-transcript-txt/scripts/video_transcript_txt.py \
"/absolute/path/to/video.mp4" \
--output "/absolute/path/to/result.txt"
Select a model, language, or cache directory
uv run ~/.claude/skills/video-transcript-txt/scripts/video_transcript_txt.py \
"/absolute/path/to/video.mp4" \
--model small \
--language zh \
--model-dir "~/.cache/whisper"
Supported operational options include --model (default small), --language (default zh), --model-dir (default ~/.cache/whisper), and --force. Use --force only after explicit user authorization to overwrite an existing final TXT.
Video Transcript TXT Data Schema & Taxonomy
Input and output organization
| Item | Format or rule |
|---|---|
| Input | One or more absolute paths to complete local video files |
| Default output | <video filename>_整理版.txt in the video's directory |
| Custom output | A user-supplied .txt path via --output |
| Final transcript | One TXT file per successfully processed video |
| Source media | Never deleted, moved, or modified |
| Intermediate files | Whisper JSON and related outputs exist only in a system temporary directory and are always cleaned up |
Transcript line format
Each valid segment is written in chronological order using:
[HH:MM:SS.mmm - HH:MM:SS.mmm] text
The transcript does not invent speaker identities or silently correct names, numbers, dosages, or technical terms based on context.
Machine-readable metadata
The terminal emits a SUMMARY_JSON: record for automation. Per-video summary fields include:
- Source video absolute path
- Final TXT absolute path
status:successorerror- Failure reason, when applicable
- Video duration in seconds
- Valid segment count
- Actual single-video processing time in seconds
- Whisper model and language
For batch runs, the aggregate report tracks successful and failed counts, each output path, each video's duration, each processing time, total duration of successfully processed videos, and wall-clock elapsed time for the complete batch. These two duration measurements remain explicitly separate.
Video Transcript TXT Advanced Features
- Timestamp precision down to milliseconds for every retained speech segment.
- Whisper model selection through
--model, withsmallas the default. - Language selection through
--language, defaulting tozhwhen unspecified. - Configurable Whisper model caching through
--model-dir. - Safe overwrite controls that require explicit authorization before
--forceis used. - Atomic final-output replacement to reduce the risk of incomplete TXT files.
- Temporary-directory isolation and cleanup on both successful and failed runs.
- Batch orchestration through a main agent and one-video-per-sub-agent execution.
- Concurrency limits of two videos by default and up to three when resources and video lengths permit.
- Graceful serial fallback when sub-agents or parallel execution are unavailable.
- Failure isolation: one failed video does not block unrelated videos in the same batch.
- Machine-readable
SUMMARY_JSON:output for downstream agents, scripts, and Openclaw Skills integrations. - Clear distinction between aggregate media duration and wall-clock processing time during parallel execution.
SKILL.md
Loading
Related Openclaw Skills
A-Share Index Futures Position Statistics - CFFEX
A Python-based Openclaw Skills tool that downloads official CFFEX member-position XML data and produces validated A-share index futures position reports.
Rilke, Rainer Maria — Literary Voice
A literary voice skill that guides AI toward Rainer Maria Rilke’s concentrated attention, spiritual intensity, and object-centered poetic style.
Dutch Bros Menu Research
A research skill that helps AI agents answer accurate, practical questions about Dutch Bros drinks, prices, nutrition, availability, and customization.
Time as Aesthetic Subject
A cross-domain aesthetic framework that makes time, mortality, and transience visible in creative work.
JoinQuant Strategy Backtest Code Generator
A JoinQuant-focused Openclaw skill that generates ready-to-run quantitative strategy backtest code for multifactor stock selection and technical signal timing.
Jira Integration Skill
A Jira API integration skill that automates issue creation, status synchronization, and conditional workflow operations for AI agents.
Featured*
