Transcript gist

Extract the editorial substance of a transcript and keep useful source timestamps.

Summarize the supplied transcript for someone deciding whether to watch or listen. Lead with the central point, then group supporting ideas by theme or chronology. Keep concrete findings, qualifications, and disagreements. Use only the available transcript; do not infer missing sections.

Omit sponsor reads, promotions, repeated introductions, and calls to subscribe, unless the user specifically asks to analyze them. Do not spend summary space announcing those omissions. Keep editorial discussion about a company or product when it matters to the topic.

Use short paragraphs or a small set of bullets. If timestamps are supplied, add a brief Key moments list using exact timestamps from the source. Never invent timestamps, reactions, or quotes. Follow the requested length without padding. Return only the summary.

Example

Fictional repair-café podcast transcript

Before 160 words

[00:00] Welcome back to the Community Workshop podcast. Please subscribe and tell a friend about the show. [00:25] Today we compare the two ways our repair café booked visitors during June: reserved slots and walk-ins. We tried each on one Saturday, with four volunteer repairers at both events. [01:40] Reserved slots had a median wait of 12 minutes. Walk-ins had a median wait of 28 minutes. We counted waiting time from arrival until a repairer began. [03:10] Six people with reservations did not attend. The walk-in event had no appointments to miss, but three people left before a repairer was free. [04:20] Our sponsor has a special offer this week. Use the code in the description. [05:05] The weather differed between the Saturdays, and we did not record how difficult the repairs were. We cannot attribute the waiting-time difference to booking format alone. [06:15] Next month we plan a mixed format, but the volunteer coordinator has not approved the schedule.

After 95 words

In June, a repair café tested reserved slots and walk-ins on one Saturday each, with four repairers per event. Median arrival-to-repair waits were 12 and 28 minutes respectively. Six reservation holders did not attend; three walk-in visitors left before service. Weather differed and repair difficulty was unrecorded, so the wait difference cannot be attributed to booking format alone. A mixed-format trial is planned for next month; its schedule is unapproved. Key moments - [00:25] Test setup - [01:40] Waiting times - [03:10] Missed reservations and walk-outs - [05:05] Comparison limits - [06:15] Proposed next trial

Examples illustrate the method. They do not measure model output.

Files and sources

prompt.txt Complete prompt as plain text.
SKILL.md The same prompt with a skill header.
llms.txt (JSON) Prompt, metadata, example, and sources.
LICENSE MIT license text.

Adapted from steipete/summarize · Transcript gist. Source notes.