Write and cut cold opens for podcast and video episodes: the 25 to 50 second hook that runs before the intro or title. Covers picking the strongest moment or claim, six structures by episode type, a timing system with word budgets, cut points, and a cold listener test. Use when an episode opens slow, when a transcript needs its best moment moved up front, when writing a teaser, pre-title sequence, or podcast intro hook, or when retention drops in the first 30 seconds.
./install.sh cold-open-craftname: cold_open_craft description: Write and cut cold opens for podcast and video episodes: the 25 to 50 second hook that runs before the intro or title. Covers picking the strongest moment or claim, six structures by episode type, a timing system with word budgets, cut points, and a cold listener test. Use when an episode opens slow, when a transcript needs its best moment moved up front, when writing a teaser, pre-title sequence, or podcast intro hook, or when retention drops in the first 30 seconds.
A cold open is the scene, claim, or exchange that runs before the title sequence or host intro. It jumps straight into the story with no greeting, no logo sting, and no setup. Its job is narrow: give a new listener one specific reason to stay, and leave one question open that only the episode can answer.
This skill is a timing and selection system, not a prompt for a catchy first line. A good first line inside a 90 second ramble is still a failed cold open.
Default target: 25 to 50 seconds.
Word budget comes from speaking rate. The National Center for Voice and Speech puts average conversation at about 150 words per minute, with radio hosts and podcasters at 150 to 160 and presentations at 100 to 150. At 150 words per minute:
At a slower 130 words per minute, 25 to 50 seconds is about 54 to 108 words. Measure the actual speaker when you can: time a one minute sample from the recording and count its words, then use that rate. Run drafts through the checker:
python3 ~/workspace/skills/cold-open-craft/bin/cold-open-check.py --file draft.txt --wpm 150
Exceptions, stated plainly: a video Short or a cold open under 25 seconds can work, and television teasers often run longer. The 25 to 50 second window is this skill's working target for podcast and online video episodes, because YouTube measures its Intro metric at the first 30 seconds and a cold open that is still running at 90 seconds has become the episode, not the open.
Work from the transcript or recording, not from memory of the episode. Pull every candidate that has at least one of these:
Write each candidate down with its timestamp and its exact words. If you cannot quote it, you do not have it yet.
Score each candidate 0, 1, or 2 on five tests. Keep only candidates at 8 or above. Drop any candidate that scores 0 on test 1 or test 2, whatever its total.
Full shapes and worked examples are in references/structures-and-examples.md. Short version:
Television craft names three basic kinds, standalone, episode setup, and teaser that moves a later event up front. The six shapes above are those three kinds fitted to podcast and video episodes.
Lay the chosen cold open out in beats before writing or cutting a word:
If beat two needs more than two facts to make sense, the candidate failed the standalone test. Pick a different candidate instead of adding explanation.
For a recorded moment: cut tight. Remove greetings, false starts, and any sentence that repeats what the previous sentence said. Keep one natural pause before the final beat if the speaker took one. Do not cut so tight that the speaker sounds rushed.
For a written open: draft to the word budget, read it aloud with a timer, and cut to the spoken time, not the page count. Reading speed on paper lies.
Run the checker on the final text. Fix every failure it reports, then do the human test in step 7. The checker catches length and throat clearing. It cannot judge whether the moment is interesting.
Play or read the cold open, and nothing else, to someone who knows nothing about the episode. Ask four questions:
Pass marks: they name the topic in one line, they name a specific thing they want answered, and they report no drift. If question 2 gets a shrug, the open has no loop. If question 1 gets the wrong topic, the open fails promise match, even if it is exciting.
No spare listener available: read it aloud yourself after a break, and answer the four questions in writing from the text alone. Weaker than a real listener, better than skipping the test.
A cold open teases. It does not deliver.
For YouTube: open the audience retention report in YouTube Studio. YouTube defines Intro as the percentage still watching after the first 30 seconds, and flags spikes for rewatched or shared moments and dips for skipped or abandoned moments. Compare this episode's Intro percentage with your own recent episodes of similar length, not with a number from someone else's channel. A dip inside the cold open itself points at the exact beat to cut or move next time.
For audio podcasts: use the consumption or completion data your host provides, compare the episode against your own recent episodes, and note where you changed the open. Keep a one line log per episode: structure used, length in seconds, and the result. Ten episodes of that log teaches more than any general rule.
When asked to write or fix a cold open, return:
publisher key 2e43c913833850e6d80f939677e3de76e806d47785c0f1caaca90f9d18c98e1e
No ratings yet — be the first to rate it.