transcription-helper

Guides users through video transcription workflow from input to output. Transcribes local video files and YouTube URLs using gpt-4o-transcribe. Use when users want to transcribe videos, audio files, YouTube content, or need help with media-to-text conversion.

$ Installer

git clone https://github.com/costiash/CognivAgent /tmp/CognivAgent && cp -r /tmp/CognivAgent/app/agent/resources/.claude/skills/transcription-helper ~/.claude/skills/CognivAgent

// tip: Run this command in your terminal to install the skill


name: transcription-helper description: Guides users through video transcription workflow from input to output. Transcribes local video files and YouTube URLs using gpt-4o-transcribe. Use when users want to transcribe videos, audio files, YouTube content, or need help with media-to-text conversion.

Transcription Helper

Entry Points

This skill can be invoked at different stages:

Entry PointWhenStart At
New transcriptionUser wants to transcribe videoPhase 1
Job completedBackground transcription job finishedPhase 4
Resume workflowUser returns to a saved transcriptPhase 4

Job Completion Flow: When a transcription job completes, the system automatically requests Phase 4 to present results and options to the user.

Workflow Phases

Phase 1: Gathering Input

  1. Greet briefly (mention you use gpt-4o-transcribe for high accuracy)
  2. Ask for:
    • Video source (local file path or YouTube URL)
    • Language (optional — e.g., 'en', 'es', 'zh' — auto-detects if not specified)
    • Domain vocabulary (optional — technical terms, proper nouns to improve accuracy)
  3. Keep it concise. Don't overwhelm with detailed explanations.

Phase 2: User Confirmation

ONLY proceed after explicit confirmation ("yes", "proceed", "confirm", "go ahead"):

  • If changes requested → return to Phase 1
  • If confirmed → proceed to Phase 3

Phase 3: Transcription

  1. Use transcribe_video with:
    • video_source: File path or YouTube URL
    • language: ISO 639-1 code if known (e.g., 'en', 'es', 'zh')
    • temperature: 0.0 for consistent results
    • prompt: Domain vocabulary if provided
  2. The tool creates a background job and returns immediately with a job ID
  3. Tell user to monitor progress in the Jobs panel
  4. DO NOT call save_transcript — the job automatically saves the transcript when complete
    • YouTube videos: Title is auto-extracted from yt-dlp for evidence linking in KG
    • Local files: No title is extracted (title will be None)
    • The transcript is registered with a unique ID automatically

IMPORTANT: When the job completes, the system triggers Phase 4 directly. The transcript is already saved — proceed to show results, do NOT call save_transcript again.

Phase 4: Results & Follow-up

After successful transcription:

  1. Report completion and share transcript ID
  2. Show preview (~200 characters)
  3. Share metadata (source type, length, splitting info)
  4. Present 5 options:
OptionDescription
1. SummarizeCreate concise summary with key points
2. Extract Key PointsList main topics and actionable items
3. Show FullDisplay complete transcription
4. Save Derived ContentSave summary/notes using content-saver skill
5. Build Knowledge GraphExtract entities and relationships (recommended for rich content)

Ask: "What would you like me to do with this transcription? Choose 1-5, or describe something else."

Option 4 Flow: When user selects "Save Derived Content":

  1. First generate the content to save (summary, notes, key points)
  2. Invoke content-saver skill for format selection
  3. The skill handles format templates, filename suggestions, and file saving

Error Recovery

Error TypeTroubleshooting
YouTube errorsCheck URL validity, video availability, age restrictions
File errorsVerify path exists and is valid video format
FFmpeg errorsEnsure FFmpeg is installed
API errorsCheck OPENAI_API_KEY is set correctly
Timeout errorsVideo may be too long; suggest splitting

Repository

costiash
costiash
Author
costiash/CognivAgent/app/agent/resources/.claude/skills/transcription-helper
4
Stars
0
Forks
Updated2d ago
Added6d ago