| name | Video Transcript Analyzer |
| description | Analyze customer interview transcripts (SRT or plain text) to generate thematic breakdowns with summary, quotes, topics, timestamps, and full transcript. Use when given video transcripts or asked to create chapter markers. |
| version | 2.0.0 |
| allowed-tools | ["Read","Write"] |
Video Transcript Analyzer
Overview
Analyze video transcripts from customer interviews, user research sessions, or feedback calls. Produces a comprehensive structured document with thematic analysis, key insights, chapter markers, and the complete original transcript.
When to Apply
- User provides a video transcript file (.srt or .txt)
- User asks to "analyze this customer call/interview"
- User requests "chapter markers" or "topical breakdown" from a transcript
- User wants to extract insights from a recorded conversation
Do NOT use for:
- General text summarization (not conversation transcripts)
- Meeting notes that aren't transcripts
- Live conversations (only recorded/transcribed ones)
Workflow
Step 1: Detect Format
Read the transcript file and determine SRT vs plain text format.
Read: instructions/format-detection.md for detection rules.
If no video URL found in user context, prompt for it before proceeding.
Step 2: Collect Metadata
Prompt the user for customer name and company.
Read: instructions/naming-conventions.md for filename format and prompting rules.
Step 3: Analyze Themes
Read the entire conversation and identify 5-10 major topics.
Read: instructions/theme-analysis.md for grouping rules and per-theme structure.
Key rule: group thematically, NOT chronologically.
Step 4: Extract Quotes
Select 5-8 verbatim quotes from the customer/interviewee.
Read: instructions/quote-extraction.md for selection criteria and critical rules.
Key rule: quotes must be EXACTLY as spoken. Zero paraphrasing.
Step 5: Write Call Summary
Write 2-3 paragraphs covering:
- Who was on the call and their role
- What was discussed at high level
- Main themes that emerged
- Key outcomes or next steps
Keep it concise — someone should be able to skip the rest and still understand.
Step 6: Assemble Document
Use the output template in templates/analysis-output.md to structure the final document.
Output filename follows instructions/naming-conventions.md.
Step 7: Run Advisory Board Review
Before finalizing, run the draft through the three personas in eval/advisory-board.md:
- The Product Researcher — checks no insights were lost
- The Quote Verifier — ensures verbatim fidelity
- The Structure Editor — evaluates readability and flow
Fix any issues found. Quote Verifier findings are zero-tolerance.
Step 8: Validate Against Checklist
Run through every item in eval/checklist.md. All items must pass before delivering.
Step 9: Save and Report
Save the document to the appropriate location. Confirm with the user:
- File path and name
- Number of themes identified
- Number of quotes extracted
Related Skills (Workflow Chain)
┌─────────────────────┐
│ wistia-uploader │ → Upload video, get transcript
└──────────┬──────────┘
│
▼
┌─────────────────────────────┐
│ video-transcript-analyzer │ ← YOU ARE HERE
│ (analyze transcript) │
└──────────┬──────────────────┘
│
┌─────┴─────┐
▼ ▼
┌──────────┐ ┌───────────────────────────┐
│ video- │ │ interview-synthesis- │
│ clipper │ │ updater │
│ (clips) │ │ (update synthesis docs) │
└──────────┘ └───────────────────────────┘
Reference Examples
- Good (SRT + full name):
examples/good/srt-with-full-name.md
- Good (plain text, no URL):
examples/good/plain-text-no-url.md
- Bad (paraphrased quotes):
examples/bad/paraphrased-quotes.md
File Map
video-transcript-analyzer/
├── Skill.md ← You are here (orchestrator)
├── instructions/
│ ├── format-detection.md ← SRT vs plain text rules
│ ├── quote-extraction.md ← Verbatim quote selection
│ ├── theme-analysis.md ← Grouping and structure rules
│ └── naming-conventions.md ← Filename format and prompts
├── templates/
│ └── analysis-output.md ← Output document template
├── examples/
│ ├── good/
│ │ ├── srt-with-full-name.md ← SRT with timestamps
│ │ └── plain-text-no-url.md ← Plain text, no video
│ └── bad/
│ └── paraphrased-quotes.md ← Anti-pattern: cleaned-up quotes
├── eval/
│ ├── checklist.md ← Pass/fail validation
│ └── advisory-board.md ← 3 reviewer personas
└── resources/
├── EXAMPLES.md ← [legacy — see examples/]
└── REFERENCE.md ← [legacy — see instructions/]
Security and Privacy
- Handle customer names and company information with care
- If transcript contains sensitive data (credentials, PII beyond names), warn user
- Do not modify or sanitize quotes unless explicitly requested
- Preserve all content from original transcript
