Skip to main content

clip-finder

Surface the 5-8 most clippable moments in an episode with exact timestamps and a scroll-stopping hook for each.

Informations de source

Dépôt
ryanfrigo/clawmart
Dernière activité de la source
3 juillet 2026 à 06:45
Langue détectée de SKILL.md
anglais
Étoiles
3
Forks
0

Options d'installation

Le prompt qui vérifie d'abord la source est sélectionné par défaut. Vous pouvez passer à une commande directe ou télécharger une copie locale.

Vérifiez les fichiers source

Lisez SKILL.md et les fichiers associés affichés par SkillsMP avant de décider de l'installer.

Affichage de SKILL.md

SKILL.md
Instructions source · Aperçu en lecture seule
name
clip-finder
description
Surface the 5-8 most clippable moments in an episode with exact timestamps and a scroll-stopping hook for each.
metadata
{"openclaw":{"emoji":"✂️"}}
# Clip Finder Finds the short-form gold buried in a long recording. Reads a clean transcript, scores every moment against what actually performs on short-form, and returns 5–8 self-contained clips with exact in/out timestamps and a hook line for each. This is where "one recording, a week of content" starts. ## When to use (trigger phrases) Use this skill when the user says: - "find the best clips in this episode" - "what should I cut for Shorts / TikTok / Reels?" - "pull the most viral moments" - "give me timestamps for clips" - "where are the good soundbites?" ## How it works 1. **Read the canonical transcript** (`transcript.clean.md`; run `transcript-clean` first if needed). You need timestamps — if the source had none, tell the user your in/out points are estimates. 2. **Score moments against a rubric.** A moment is clippable when it hits one or more of these — the more, the better: - **Contrarian take** — "Everyone says X; that's wrong because…" - **Concrete story** — a specific, named, time-bound anecdote (not a generality). - **Emotional peak** — surprise, conviction, frustration, a laugh. - **Quotable one-liner** — a sentence that stands alone without context. - **Actionable tip** — a do-this-now instruction with a clear payoff. - **Tension / disagreement** — host and guest pushing on each other. - **Surprising data or reframe** — a number or mental model that flips the listener's assumption. 3. **Pick 5–8 non-overlapping segments.** Each must be **self-contained** — it makes sense to someone who never heard the rest of the episode. Target **20–90 seconds** (15–45s for TikTok/Shorts/Reels, up to 90s for a LinkedIn/X video). Reject anything that needs "as I said earlier" to land. 4. **Set tight in/out points.** Start on the *payload*, not the wind-up — cut the "yeah, so, I think what's interesting is…" preamble. End on the punch line or the period, never mid-thought. Give the trimmed start/end timestamps, not the raw paragraph bounds. 5. **Write a hook for each.** The hook is the on-screen text / opening line that stops the scroll. Make it a promise, a tension, or a bold claim drawn from the clip — never clickbait the clip can't pay off. 3–8 words is the sweet spot for an overlay. 6. **Tag each clip** with: the verbatim quote, why it works (which rubric items), best platform(s), and a one-line caption. Rank the list best-first so the user cuts the top 2–3 even if they're short on time. 7. **Hand off for cutting.** Output the timestamps in a form a video tool can use. If the user wants, provide an `ffmpeg` command per clip (see Notes) — but the actual cut happens in their editor, not here. ## Output A ranked list. Example (two of six shown): ```markdown ### 1. "You don't have a churn problem" — 00:22:40–00:23:18 (38s) **Hook:** "Your churn problem is a lie" **Quote:** "You don't have a churn problem, you have an empty-state problem. Nobody leaves a product that's already full of their stuff." **Why it works:** contrarian + quotable one-liner + reframe. Stands alone. **Best for:** Shorts, Reels, X video. **Caption:** The retention reframe most founders miss. Full episode in bio. ### 2. The 3am pager story — 00:31:05–00:32:20 (75s) **Hook:** "The night we deleted prod" **Quote:** "It's 3am, I'm half asleep, and I run the migration on the wrong database…" **Why it works:** concrete story + emotional peak. Great for LinkedIn video. **Best for:** LinkedIn, X. **Caption:** A hard-won lesson about backups. ``` Followed by an optional cut sheet: `Clip 1 → in 00:22:40 out 00:23:18 · Clip 2 → in 00:31:05 out 00:32:20 …` ## Notes - **OpenClaw finds the moments and writes the hooks — it does not cut video.** Hand these timestamps to your editor: Descript, Opus Clip, CapCut, Premiere, or an ffmpeg-based skill. If a clip is off by a beat, nudge it in the editor; ASR timestamps drift by a second or two. - **Optional ffmpeg cut** (requires `ffmpeg` installed by the user): to lift a clip losslessly, `ffmpeg -i episode.mp4 -ss 00:22:40 -to 00:23:18 -c copy clip1.mp4`. For vertical 9:16 with re-encode, crop/scale in a second pass. This is a convenience, not the product — verify the cut. - **The hook is a promise the clip must keep.** Don't write a hook the segment can't pay off — it tanks watch-time and trust. No fake stats, no "this one trick." - **Self-contained beats "best line."** A brilliant sentence that needs three minutes of setup is a worse clip than a good sentence that stands alone. - Spot-check the top clips against the actual audio before publishing — tone and timing don't fully survive text.
Voir sur GitHub