clip-map

# Clip Map

A long episode is a mine, not a post. Find the moments that stand alone, stamp the clocks, write the hook line each cut earns, and stop before you edit.

Inspired by SkillMedev creator-studio content-repurposing / podcast-to-clips concepts (MIT). Original Grok remake. Complements beat-script (writes the long script), hook-3s (rewrites each clip's first 3s), thumb-pack (thumbnail for each), cap-cue (captions), shot-to-cut (assemble/render), motion-craft (polish). Real video planning, not a GIF converter.

Read `references/clip-rails.md` before you map. Fill the three templates under `templates/`.

## When to use

- Podcast episode, interview, keynote, webinar, or long YouTube talk → short-form clip candidates
- "Turn this into Shorts / Reels / TikToks," "cutdowns," "clip sheet before editing"
- Need 5–12 candidates with in/out guesses, a hook line, why it stands alone, and a platform note
- Planning which moments deserve a cut *before* anyone opens the editor

## When NOT to use

- Writing the long-form spoken script / YouTube first-30s retention beats → use beat-script
- Rewriting only the first 3 seconds of an already-chosen clip (spoken + frame-one + muted overlay) → use hook-3s
- Caption / SRT / VTT track → use cap-cue
- Thumbnail + title click package → use thumb-pack
- Assemble, concat, render QA → use shot-to-cut
- Burn-in, kinetic type, motion polish → use motion-craft
- Full multi-format repurposing (threads, newsletters, quote cards as the main ask) — this skill maps *video clips*, not a week of every channel
- GIF / still-to-video converters
- Phishing, malware pitches, fake medical cures, guaranteed-return finance, CSAM, sexual content involving minors

## Inputs

Gather what you can. Infer missing pieces and LABEL guesses. Ask at most two questions if critical data is missing.

- Source: link, file name, transcript, show notes, or chapter list. Prefer transcript + chapters when available.
- Runtime and format (podcast audio-only, A/V interview, stage talk, screen-share webinar)
- Audience and primary short-form platforms they actually post to (TikTok / Reels / Shorts / LinkedIn native video)
- Moments they already named ("the story about X," "the pricing rant," chapter titles). Do not invent quotes they did not say.
- Target clip count (default 5–12) and preferred length band (e.g. 30–60s, 60–90s)
- Any hard bans (guest name spellings, topics to skip, brand-safe rules)

## Procedure

1. Read `references/clip-rails.md`. Refuse phishing, malware, fake cures, guaranteed-return finance, CSAM, and sexual content involving minors. Do not invent speaker quotes. Mark guessed clocks `[TIME: guess]`.
2. Fill `templates/source-card.md`. Lock runtime, format, audience, platforms, and every moment they named. Say what evidence you have (transcript / chapters / memory of a listen / none).
3. Audit for standalone moments. A clip candidate must work with zero surrounding context: clear setup inside the cut, a payoff inside the cut, and a reason a cold viewer would stay. Ban "part of a longer point" slices.
4. Aim for 5–12 candidates. Fewer than 5 only if the source is thin; more than 12 only if they asked and every row still passes the standalone test. Prefer quality over calendar filler.
5. For each candidate on `templates/clip-sheet.md`: working title, in/out with `[TIME: guess]` when not measured, estimated duration, hook line (what cold-opens the cut — not a full hook-3s pack), why it stands alone, best platform + length band, and any edit notes (jump the throat-clear, keep the laugh, crop to speaker).
6. Rank lightly: star 2–4 "cut first" rows so the editor is not staring at a flat list. Do not invent rankings from vanity — star the ones with the sharpest claim or story.
7. Fill `templates/handoff.md`: which clips go to hook-3s next (opener rewrite), which need thumb-pack (especially Shorts/YouTube), which need cap-cue, which are ready for shot-to-cut assembly, and which can wait for motion-craft polish. One line per hand-off.
8. If you have no transcript and no chapters, ask at most two questions (runtime? any moments they already love?), then map from what they describe and mark every clock `[TIME: guess]`. Never fake a confident timeline.

## Output format

- Source card (runtime, format, audience, named moments, evidence, labeled guesses)
- Clip sheet (5–12 candidates: in/out, hook line, standalone why, platform)
- Handoff card (hook-3s / thumb-pack / cap-cue / shot-to-cut / motion-craft next steps)
- One line: clip planning sheet; not a long script, not captions, not a render

## Safety

- Follow `references/clip-rails.md` with no exceptions.
- No phishing, malware pitches, fake medical cures, or guaranteed-return finance clips.
- No CSAM or sexual content involving minors.
- Do not invent quotes, admissions, or numbers the speaker did not say. Paraphrase only when labeled as paraphrase; prefer short attributed lines from the source.
- Mark guessed timestamps `[TIME: guess]`. Never present a guessed in/out as measured.

## Stop conditions

- Three templates delivered, or refused for scam / wrong job (long script → beat-script; first-3s only → hook-3s; captions → cap-cue; thumbnails-as-main → thumb-pack; assemble → shot-to-cut; GIF converter) / no source and no way to proceed