v1.3.03 loop iterations

Transcript to Quotes

Most transcripts contain 2-5 genuinely shareable moments buried in thousands of words. The difference between a quote that gets 50 impressions and one that gets 50,000 isn't the idea — it's whether the line works ripped out of context.

What this skill does

Not every interesting sentence is a shareable quote. Shareable quotes have specific properties — self-contained, opinionated, concise, surprising, emotionally resonant — and most of them are buried in conversational filler that hides the line until you know what to look for. This skill scans the full transcript for moments that shift your understanding or name an emotion the audience hasn't articulated, then scores each against the five qualities. A line needs at least three of the five to make the cut.

It's honest about what it skips. Generic motivational language ("believe in yourself") gets discarded. Inside references that need episode context to land get discarded. Statements that are factually interesting but emotionally flat get discarded. Lines that depended on vocal delivery to work — sarcasm, dramatic pauses — get discarded even if they were the best moment in audio, because they fall flat as plain text on a social card. The output is fewer quotes than you might expect and stronger than what extraction-by-vibes produces.

Editing ethics are explicit. Verbatim quotes get labelled "Verbatim". Light edits (filler removed, redundant clauses tightened) get labelled "[Edited for clarity]" with the original in parentheses so the user can approve. Paraphrased versions get flagged separately and used sparingly — only when the idea was brilliant but the delivery was messy. Misquoting a guest destroys trust permanently, so the skill never changes meaning or adds words the speaker didn't say.

You get three tiers — Hero (2-3), Strong (3-5), Supporting (3-4) — so you know which to lead with and which to drip-post as fillers. Bonus deliverables: the single best quote reframed as the opening tweet of a thread, an audiogram recommendation with start/end timestamps, and a five-quote carousel set arranged as a coherent arc with a suggested title slide. Quotes are pulled from across the full transcript, not clustered in one ten-minute segment, so the resulting content calendar stays varied over weeks of posting.

When this triggers

·You have a podcast, interview, or talk transcript and want the shareable moments pulled
·You need social cards, audiogram clips, or pull quotes from a longer recording
·Your current quote extraction misses the bangers and surfaces the obvious ones
·You want quotes graded by tier so you know which to lead promo with
·You need a context-flag on any quote that could be misread out of context

Example

Trigger

User pastes a 9,000-word interview transcript and asks for shareable quotes for social cards + an audiogram.

Output

Tier 1 — Hero quotes (2): 1. "Most marketing advice is optimised for the person giving it, not the person receiving it." — Jane Doe (verbatim, 14:22) Why: opinionated, surprising, 14 words, screenshot-ready. Visual: dark card, bold serif, headshot bottom-right. 2. "I stopped charging for my time the day I realised nobody buys time. They buy outcomes." — Jane Doe (32:08) [Edited for clarity. Original: "...the day I, like, realised that, you know, nobody buys time..."] Tier 2 — Strong quotes (4). Tier 3 — Supporting (3). Audiogram pick: 41:10-42:05, the failure story. Self- contained, dramatic, ends on the lesson. Carousel set (5 quotes) arranged as: problem → reframe → method → result → CTA. Title slide: "What pricing got wrong (and how she fixed it)."

Get this skill + 4 more

Included in the The Creator Stack — create more, better, faster. Save $120+ vs buying individually.

Get The Creator Stack — $99

What you get

113-line SKILL.md, ready to drop into ~/.claude/skills/
Tested through 3 Karpathy-loop iterations (versions v1.0.0 → v1.3.0)
Triggers automatically when relevant — no command to remember
Lifetime updates as the skill is refined further