format-script-to-style

format-script-to-style

Purpose

Take a delivery style from DELIVERY_STYLES.md and a target duration, and transform the script currently in the calling agent's conversation context into a v3-tagged script ready for voice generation. The output should match the chosen style's beat structure, signature language, pacing, and audio-tag recipe.

Inputs

• delivery_style_name: a key from DELIVERY_STYLES.md (e.g. UGC-Creator-Confiding).
• target_duration_seconds: an integer; the desired final spoken length after any post-processing (e.g. 29 for a typical UGC ad).

Assumed context (not loaded by this skill): the unformatted script content lives in the calling agent's conversation. This skill does not accept a --script flag or read from a file — the calling agent supplies the content from its own context and applies this skill's rules to it.

Workflow

1. Load the style entry for delivery_style_name from /Users/0xhbam/Desktop/Cursor/content-goose/DELIVERY_STYLES.md. If the name is not found, list available styles and stop.
2. Compute target word count. Use the style's word-count-by-duration table if available, otherwise derive from target_duration_seconds × (raw_wpm / 60). For styles that apply atempo > 1.0 post-process, compute on the rendered (post-atempo) duration, not the raw v3 read.
3. Map the existing script to the style's beat structure. The agent rewrites the script so each beat lands close to its template, preserving the brand/product facts already in conversation context. Hit each beat's word-count target within ±20%.
4. Apply v3 tagging:
   • Open with the single directional audio tag specified in the style's recipe (e.g. [soft, casual]).
   • Insert … for breath-pauses inside the slower beats named in the recipe.
   • Insert — for emotional pivots in the beats named in the recipe.
   • Do not stack or swap audio tags mid-script — they re-color all subsequent sentences.
5. Apply signature language patterns from the style entry (softeners, possessive intimacy, sensory triads, signature close).
6. Avoid anti-patterns listed in the style entry.
7. Return the tagged script as plain text (paragraph per beat). Include a comment block at the bottom listing target duration, target word count, actual word count, and the voice ID + settings recommended by the style entry — so the next skill (generate-voiceover-elevenlabs-v3) has everything it needs.

Output

Tagged script as text, returned to the calling agent. Example shape:

[soft, casual] <Beat 1 line with … breath pauses>

<Beat 2 line with — pivot>

...

<!-- format-script-to-style
delivery_style: UGC-Creator-Confiding
target_duration_seconds: 29
target_word_count: 115
actual_word_count: 118
recommended_voice_id: jqcCZkN6Knx8BJ5TBdYR  # Zara
voice_settings: {stability: 0.40, similarity_boost: 0.75, style: 0.05, use_speaker_boost: true}
post_process: atempo=1.2
-->

Quality Checks

• Every beat from the style's structure is present in the tagged script.
• Actual word count is within ±20% of target word count.
• Exactly one directional audio tag appears at the open. No additional audio tags mid-script.
• … and — are placed only in the beats specified by the style's recipe.
• Signature close is preserved verbatim if the style entry specifies one (e.g. "perfect for every day").
• None of the style's anti-patterns appear in the tagged script.

Failure Modes

• delivery_style_name not found in DELIVERY_STYLES.md — list available, stop.
• Calling agent has no script content in context — surface this and stop; this skill cannot invent product facts.
• Target duration is unachievable for the style (e.g. asking for 10s of a style whose minimum is ~24s). Surface the conflict and ask for either a longer duration or a different style.
• Script content involves claims/CTAs forbidden by the style's anti-patterns and cannot be reformulated without losing the brand message — surface the conflict to the user.