Pick clips from an episode

Find moments a stranger can enjoy and understand. Give the editor clear cuts and the team a reason to post each one. Connect the clip to more of the show without withholding its own payoff.

Use this skill to pick moments and brief their edits and release. If the user asks for finished video files, use available editing tools and check the exported video. If those tools are absent, say that the output is an edit brief.

Get enough context

Read SHOW-PROFILE.md in the show’s working folder if it exists. It’s a short notes file about the show (audience, look, past titles, and lessons) that the Podcasting skills share. Use its audience, look, formats, and clip history. If absent, keep a short working note: audience, show promise, format, look, recent titles, and known results. Mark guesses. Continue without a full profile.

Find the target platforms and what success means. A clip may help new viewers find the show, earn a full-episode start, or serve current fans. Pick a main job for each clip. If the user gives no goal, use reaching likely full-episode viewers as the working goal.

Input	Action
Episode link only	Seek the transcript and view or hear candidate sections if tools allow. If access fails, ask for the transcript or recording. Give a selection plan, with no invented moments or times.
Timed transcript	Keep its source name and version. Find candidates, then mark sound and visual quality as unverified until checked against the recording.
Untimed transcript	Use exact first and last lines plus section labels. Give rough lengths as estimates. Do not make up timestamps.
Notes only	Give candidate topics and questions for the editor to find. Label them as leads, not selected clips.
Nothing yet	Ask for an episode source and where the team plans to post. If recording has not happened, give prompts that may draw out complete stories or clear examples.

Ask about the edit budget only if it changes the plan. Use supplied constraints. Avoid a long intake before reading the episode.

Read the episode before ranking lines

Map the main sections. Note ideas that need earlier context. Check for claims the guest later changes or takes back. Read around each possible cut, including the next exchange. Keep reading until the meaning is clear.

Look for moments such as:

A belief the guest challenges, with a reason that supports the challenge.
A choice with a real cost, or a disagreement that gets resolved.
A story with a clear turn and an ending.
A useful demonstration or example people can follow.
A line viewers may quote, with enough context to keep its meaning.
A question people are asking now that this guest can answer well.

Keep humor or chemistry when those are central to the show. Reject an inside joke a stranger cannot follow. Do not choose a weak moment just because the guest is famous.

Check what people care about now

For a full selection pass, load references/moment-research.md. Use it to check current debate, search interest, and comparable clips. If you can’t open it, continue with the steps here. For a narrow cut review, check only the facts or timely claims that could change the verdict.

If the Benchmark MCP is connected, use it to pull real public outlier episodes, titles, and thumbnails from comparable shows. Otherwise use public pages, search, and user evidence. Keep dates and links in the task’s research note. Never require the connector.

Tie each timely candidate to a real audience question and a source checked today. Give it a review-by date if its value may fade. Prefer a lasting idea when the timely frame adds little. If browsing is unavailable, mark current interest as unverified and choose from the episode’s strength.

Apply the cut gates

Reject or repair a candidate before scoring if:

The cut changes the guest’s meaning or drops a needed caveat.
A stranger cannot tell who or what it refers to.
The question never gets an answer, or the story has no turn or end.
It attracts people who have little reason to want this show.
The recording cannot support a usable cut, unless a clear repair is possible.

Keep an important correction in the clip even if it makes the clip longer. For health, money, or other serious claims, check the claim before making it the hook. A guest’s title is not proof that every claim is true.

After those checks, rank each possible clip High, Medium, or Low on hook strength, payoff, audience fit, and signs of interest. Mark interest as unknown when needed. Explain the strongest and weakest point. Do not predict a viral score or a view count.

Use three verdicts: Cut, Repair, or Skip. State the exact repair. Do not fill a quota with weak moments. Return no picks when the source has none that pass.

Score moments with a decision model

For a long episode or many episodes, score every candidate moment with a decision model (for example Jev) instead of reading each one yourself. Ask fixed questions with set answers, such as:

Does it make sense without the sentence before it?
Does it start or end mid-sentence?
Does it hold one idea, with its setup and payoff?
Would a stranger stop for it?

Clip the moments with the highest scores that pass the cut gates. Then write several on-screen text hooks per clip and score those too, for example: does it say the clip’s best fact, would a stranger stop, does it sound like a person.

Run the work in nine stages:

Time the words. Get a word-level transcript. Map each sentence and the pause before it. Every later cut starts and ends at a sentence boundary with a pause.
List candidates. List every sentence span in your length range. Drop spans that start in the intro or cold open, start on a clause that needs the sentence before it, or start with no pause. Keep at most two spans from each group of overlapping spans. Spread them across the episode and group them by topic.
Note the shot. Sample a few frames from each candidate. Record how many faces show, how large they are, and how much of the frame is a screen.
Score and select. The model answers the fixed questions. Gates drop a moment. Quality scores rank the rest. Keep one moment from each overlap group, send screen-heavy moments to the wide format only, and cap the picks for each topic.
Write hooks. Write several hooks for each clip, each on a different opening method. Gate, score, and rank them. The writer may move the start to a stronger line.
Find a clean start. Try the nearby sentence starts. A start passes when the opener rules hold, a pause comes before the first word, the first second is steady, and a face is in the crop. Use the nearest start that passes. If none passes, do not cut the clip.
Edit. Snap the cut to words. Remove leading filler. Shorten pauses by audio level, because transcripts hide some pauses. Remove “um” and “uh”. Move the captions to match. Let the shot set the format: a face crop, two speakers in a box, or the full frame for a screen.
Check the export. Check the silence share and the longest silence. The first word must come almost at once, and the end must not trail. No sound of a removed word may remain at a join. Keep words in source order except for one payoff line of 6 seconds or less at the front. Move it only when the meaning stays the same. List both source ranges in the brief. Check the frames while the hook shows.
Pick and test. A person picks the clips and hooks and makes every honesty call. Test the picks and log what changed.

Load references/scoring.md to set up the gates, the scores, and the selection math. Load references/hooks-and-text.md to write and judge hooks and on-screen text. Start the model’s questions from references/rubric-clip.md, references/rubric-hook.md, and references/rubric-onscreen.md. If you can’t open them, continue with the steps here.

Set the scores to your own goal. Views, comments from the right buyers, and sign-ups each need different picks. Check the model’s top picks against your own choices on a few clips, and adjust the questions until they agree. A person still makes the final pick.

Decide how many clips

Count clips per episode. Look for about one distinct clip for each 10 minutes of episode, then keep only the clips that pass the gates.

Episode length	Distinct clips to look for
Under 20 minutes	1 or 2
20 to 45 minutes	2 to 4
45 to 90 minutes	4 to 8
Over 90 minutes	8 to 20

These numbers are a search limit, not a quota. A small show with no clip history can start with two or three and post more when each has its own strong moment. “20 clips” can mean 20 distinct clips from a long episode, or 5 clips with 4 variants each.

Choose the format

Format	Use it when	Starting edit range
Vertical Short, Reel, or TikTok	One idea lands fast and survives a tight crop	Try 20 to 60 seconds; keep 60 to 180 seconds when the story needs it and the destination supports it
Horizontal standalone video	The viewer needs a full explanation, story, or exchange	Try 3 to 12 minutes; go longer only when the section earns the time
Episode trailer	The task is to preview the whole episode and sell its broader question	Use several clearly separated moments; label the piece as a trailer

These ranges are edit starting points. They are not platform limits or claims about an ideal length. Check current destination limits before export. YouTube Shorts can run up to three minutes under its current format rules.

Post vertical clips more often when each has its own strong moment. Set a high bar for horizontal clips on the main episode channel. Each needs its own reason to watch and a package that can stand alone. A separate clips channel can post at a different pace if viewers want it and the team has time.

Do not cut every candidate into every format. A complete four-minute answer may work as a horizontal video while its most surprising sentence fails alone.

Pick the platform first

Ask where each clip goes before you cut it. The same moment can work on several platforms in a different shape.

Platform	Shape	What to pick
Instagram, TikTok, YouTube Shorts	Vertical 9:16 with a text hook on screen. Shorter and tighter usually wins.	The fastest, clearest moment. Test hooks as trial reels.
X	Horizontal 16:9, longer, with more depth	The more technical or detailed moment
LinkedIn	Horizontal 16:9, longer, with more depth	The moment a business viewer can use at work

Do not post every clip at once. One or two clips a day per platform is a sensible start. On Instagram, test extra versions as trial reels instead of posting them to followers.

Make variants

A variant keeps the same moment and changes one thing. Then a test shows which change made the difference. Run variants as Instagram trial reels, which show a reel to non-followers first, and keep the one that performs best.

New hook: change the opening line or on-screen hook. Use it when the moment is strong but the first seconds may not stop a stranger.
16:9: keep the full horizontal frame. Use it when both speakers, a screen, or a demo must stay in view.
Full-screen 9:16: crop tight to the speaker. Use it when one face carries the moment.
Clip in a frame: place the horizontal clip inside a frame, such as a browser window, on a vertical background. Use it when a tight crop would cut the point.
Different music: change or remove the background track under the same cut. Use it to test mood. Keep the voice clear and use licensed music.
Different length: cut a short version that goes straight to the payoff and a longer version with the setup. Use it when you are not sure how much context a stranger needs.

Every variant must pass the same cut gates. A new hook is new copy, so it must not change the guest’s meaning.

Give the editor a precise brief

Load references/edit-briefs.md when making briefs or reviewing a cut. If you can’t open it, continue with the steps here. For each selected clip, provide:

Clip ID, main job, target viewer, platform, and format.
Source version, in and out points, exact first and last words, and any context that must stay.
Opening line, first shot, exact on-screen hook, and where the payoff lands.
Cut order, proposed removals, captions, crop, and any needed repairs.
A post title or caption. For a horizontal video, include a paired title and thumbnail brief.
The next thing to watch, the link route, and why a viewer would want it.

Use a cold open, meaning start with the moment itself before any show introduction. Aim to make the point of interest clear in the first one to three seconds. Keep a short host question if the answer needs it. Avoid opening on “it”, “that”, or an unknown name without context.

Use exact quotes for the guest. Mark proposed host words or text overlays as new copy. If moving a later line to the front, list both source ranges and check that the new order keeps the meaning.

Keep captions accurate and readable. Check names, numbers, and words such as “not”. Keep key text away from app buttons and faces. Use the show’s type style. Do not add movement or sound effects just to fill quiet moments.

Plan the release and next click

Load references/release-and-learning.md for a batch plan, posting schedule, or a question about views failing to become listeners. If you can’t open it, continue with the steps here. For one clip, return its next-click route and review date.

Use a working link path. For YouTube Shorts, use Related Video when the account has access. It links to a video on the same channel. Check access before promising it. For Reels or TikTok, use the profile or Story link route the account has. For a horizontal YouTube video, brief the full-episode link. Add an end screen, a clickable prompt at the end, where available.

Deliver the clip’s answer before asking for more time. Name a different question or deeper example the episode answers next. A vague “full episode out now” gives little reason to click.

Check whether clips draw the wrong viewers, repeat old posts, or waste edit time. Those are real costs to the show. Do not claim that Shorts hurt the reach of full episodes just by sharing a channel.

Return the plan

Lead with the best clip and why it earns the edit time. Include:

A ranked candidate table with source ranges, verdict, format, and key reason. Include near misses when they teach the editor what to avoid.
Complete briefs for the selected clips, ordered by priority.
A short release plan tied to the team’s capacity and the episode link.
Open checks, evidence of current interest, and what to measure next.

Use the invented Better Questions show and fictional sleep researcher Dr. Lena Park for examples. Keep example lines clearly labeled as invented. For the worked cut, see the editing reference.

If installed, use podcast-packaging for deeper testing of a horizontal clip’s title and thumbnail. Use podcast-growth-review to diagnose results across releases. The briefs and release plan above should still work when this is the only installed skill.