Summarise Transcript

Summarize the YouTube transcript at <FILE_PATH>

Instructions:
1. Read the transcript file
2. Parse timestamps (format: HH:MM:SS text) and group into 20-minute chunks:
   - Chunk 1: 00:00:00 - 00:19:59
   - Chunk 2: 00:20:00 - 00:39:59
   - Continue for remaining content

3. For EACH chunk, generate a structured summary with these sections:
   - Key points by speaker
   - Important technical insights
   - Decisions made
   - Action items
   - Unresolved questions/follow-ups

4. Write output to <OUTPUT_PATH>

Use this format:
# Summary: <video-id>

## Chunk 1 (00:00 - 20:00)
### Key Points by Speaker
### Important Technical Insights
### Decisions Made
### Action Items
### Unresolved Questions/Follow-ups

---

## Chunk 2 (20:00 - 40:00)
[repeat structure]

3. Per-Transcript Processing (done by each agent)

Parse and Chunk:

• Read the transcript content
• Parse timestamps from each line (format: HH:MM:SS text)
• Group lines into 20-minute chunks based on timestamp

Timestamp parsing:

Pattern: ^(\d{2}):(\d{2}):(\d{2})\s+(.*)$
Seconds = hours*3600 + minutes*60 + seconds

Summarize Each Chunk: Apply the summarization prompt:

You are an expert summariser of software engineering meetings. Analyse and generate a structured summary that includes:

• Key points by speaker - Summarise suggestions, technical inputs and/or concerns raised by each speaker
• Important technical insights - Highlight significant technical and strategic insights, challenges, trade-offs and/or discoveries
• Decisions made - Note any design, process or timeline decisions agreed upon
• Action items - List specific tasks assigned, including who is responsible and deadlines
• Unresolved questions/follow-ups - List open questions, blockers, or areas requiring further discussion

4. Output Format

Create output/summary-<video-id>.md in the output directory alongside the transcript.

Example Usage

Single transcript: User: "Summarize output/transcript-Ta5g-OxjPO4.md" -> Process that one file, output to output/summary-Ta5g-OxjPO4.md

All transcripts in directory: User: "Summarize all YT transcripts" -> Glob for output/transcript-*.md, spawn parallel general-purpose agents for each