say-to-her

📁 bakabird/saytoher 📅 6 days ago
1
总安装量
1
周安装量
#44571
全站排名
安装命令
npx skills add https://github.com/bakabird/saytoher --skill say-to-her

Agent 安装分布

amp 1
opencode 1
kimi-cli 1
codex 1
gemini-cli 1

Skill 文档

Say To Her

Transform any source material into a single dramatic, entertaining conversational snippet — complete with stage directions — that a programmer can confidently deliver to his girlfriend.

Core Principles

  • ONE nugget only. Scan the source, find the single most interesting or surprising point, discard everything else. Never summarize.
  • Dramatize ruthlessly. Exaggerate, add flair, make it theatrical. Entertainment value over accuracy.
  • Stage directions are MANDATORY. Every output must include inline delivery annotations — pauses, voice cues, gestures, expressions. No exceptions.
  • Storytelling, not lecturing. Sound like an excited boyfriend sharing something wild, never like a teacher or presenter.
  • No jargon leakage. Translate all technical terms into casual language or wrap them in playful explanation.
  • No moralizing. Suppress all “this raises important questions about…” instincts. Pure entertainment.
  • Output length: 100-250 words, speakable in 30-90 seconds.
  • Output language: match the user’s prompt language. If they write in Chinese, the script is in Chinese. If English, English.
  • Default audience: smart non-technical person who is curious but has zero context on the topic.

Transformation Workflow

Transforming source material into a script follows these steps:

  1. Accept source material — pasted text, topic description, article excerpt, any format. Do NOT fetch URLs; work only with provided text.
  2. Scan for the ONE most interesting nugget — look for what’s surprising, counterintuitive, viscerally impressive, or just plain weird. Ignore everything else.
  3. Discard the rest — ruthlessly. If the source has 10 interesting facts, pick 1. Commit to it.
  4. Dramatize with storytelling structure:
    • Hook — grab attention: “So you know how…?” / “Okay, get this—” / “Wait, I read the craziest thing today”
    • Tension — build curiosity, set up the reveal, make her lean in
    • Reveal — deliver the punchline, the mind-blowing fact, the “wait WHAT?” moment
    • Closer — playful tag-out line, callback, or “right??” to land the ending
  5. Annotate with delivery stage directions throughout the script — not just at the start. See references/tone-guide.md for the full annotation catalog. Aim for roughly 1 annotation per 2-3 sentences.
  6. Format as a ready-to-use script using the output template below.

Output Template

Always format output as follows:

## Your Script

**Topic**: [one-line summary of the nugget]
**Vibe**: [emotional flavor, e.g. "mind-blown revelation", "cute fun fact", "conspiracy-theory dramatic"]
**Best delivered**: [context suggestion, e.g. "over dinner", "random text", "while cooking together"]

---

[The script with inline (delivery annotations) woven throughout]

---

**Pro tip**: [one practical delivery suggestion, e.g. "Pause after the big number — let her do the math in her head" or "If she laughs at the hook, ride that energy into the reveal"]

What NOT to Produce

Avoid these anti-patterns at all costs:

  • The TED Talk — informative but boring, no personality, sounds like a presentation. Missing: hooks, stage directions, casual language.
  • The Wikipedia Entry — accurate but dry, tries to cover everything instead of cherry-picking one nugget. Missing: storytelling arc, dramatization.
  • The Lecture — talks down to her, jargon-heavy, explains too much background. Missing: respect for the audience, entertainment value.
  • The Robot — no stage directions, flat delivery cues. The words might be good but there’s no guidance on HOW to say them.

See references/examples.md for full side-by-side comparisons of bad vs. good transformations.

Edge Cases

  • Source is already simple/casual — still dramatize. Don’t parrot; add storytelling structure and stage directions.
  • Source is very long — fine. Just pick ONE nugget from it. Length of input has no bearing on output.
  • Source is in a different language than user’s prompt — translate and transform in one step. Output in the user’s language.
  • Source has no obvious interesting nugget — find the least boring bit and dramatize it hard. If truly nothing works, tell the user and suggest different material.
  • Same source provided again — pick a DIFFERENT nugget this time. Never repeat.
  • Sensitive or controversial topic — transform it, but prepend: (read the room before telling this one) at the top of the script.

References

Load these reference files when generating output:

  • references/examples.md — Complete source-to-script transformation examples (4 good, 2 anti-patterns) demonstrating the exact tone, structure, and annotation density expected. Load this on first use or when calibrating output quality.
  • references/tone-guide.md — Full delivery annotation catalog organized by category (pauses, voice, gestures, expressions, pacing, engagement hooks) with usage guidance and combination rules. Load this when selecting and placing stage directions.