---
name: face-camera-recording-script
description: Use when the user wants a spoken video script for a recording or generated presenter, including optional teleprompter scroll padding and a separate Video Studio story-beat outline. Do not use for rendering or editing footage.
---

# Face-Camera Recording Script

Create clean spoken copy for a recorded or generated presenter video.

## Contract Checklist

- Write the script as a user-facing markdown file under the current task's
  `outputs/` directory when available.
- Keep the deliverable to a simple title and the words the presenter should
  speak.
- Do not add YAML frontmatter, source metadata, format, duration, audience,
  delivery, production notes, or other metadata blocks to the script file.
- Keep visual planning out of the script by default. Do not add bracketed
  motion-graphic, transition, caption, crop, or visual cues, and do not add an
  `Edit Notes` section unless the user explicitly requests them.
- In Video Studio, keep chapter titles in `scriptBeats` metadata, never in the spoken script or teleprompter copy.
- Keep paragraphs short so the copy is easy to perform and can be passed
  cleanly to a generated-video service.
- Add visible scroll padding only when the user needs a manual teleprompter
  script.
- Return the saved script path and mention scroll padding only when it was
  actually added.

## Workflow

1. Identify the subject and source route:
   - clarify the subject only when the user has not supplied enough material
   - determine whether the source will be manually recorded or generated by a
     presenter service
   - if the user asks for a dummy script, choose a plausible topic and continue
   - Anti-patterns: blocking on unnecessary questions, assuming manual recording, inventing unsupported source material
2. Draft the spoken copy:
   - a direct hook
   - a simple middle argument or demonstration
   - a concise closing beat
   - short line breaks suitable for reading aloud
   - use only claims supported by the user's source material
   - Anti-patterns: writing like an essay, adding visual directions, mixing production metadata into spoken copy
3. Save the clean script:
   - use a short human-readable title followed directly by the spoken script
   - omit frontmatter and metadata fields
   - keep the file under the current task's `outputs/` directory when available
   - when working on a Studio video, read `read_script` for its current text, revision, outline, and passage feedback before revising; treat unresolved comments as editorial requests, not automatic permission to expand unrelated scope
   - for a longer script or when an outline is requested, create or revise a concise sequence of story beats using `update_script_beats` with the current `expectedRevision`; each beat has a stable `id`, short `title`, and `startQuote` matching one exact unique passage in script order
   - after changing spoken text through the existing script update route, read the saved script again before updating anchors; verify `read_script` returns a resolved outline and unchanged spoken text after a beat-only update
   - if a revision conflicts, reread and reconcile instead of overwriting another person's work; if an anchor is missing or ambiguous, repair it against the current text rather than inventing a position
   - Anti-patterns: adding source or format headers, adding edit notes by default, saving only in chat, inserting chapter headings into spoken text, discarding feedback or rewriting unrelated sections to change an outline
4. Add teleprompter padding only for manual recording:
   - resolve `<SKILL_ROOT>` to this installed skill directory and use the verified Python interpreter (`py -3` or `python` on Windows, usually `python3` elsewhere). Quote paths containing spaces and use the bundled helper:
   ```bash
   <PYTHON> "<SKILL_ROOT>/scripts/add_scroll_padding.py" \
     outputs/face-camera-script.md \
     --lines 60
   ```
   - verify the bottom contains visible `.` spacer lines
   - skip this step for HeyGen or another generated-video route
   - Anti-patterns: adding padding to generated-presenter input, using blank trailing lines, using HTML breaks
5. Verify the deliverable:
   - confirm the file exists
   - confirm the spoken section contains no bracketed visual cues or default
     edit-notes section
   - confirm no frontmatter or metadata block was added
   - for Studio outlines, confirm every beat resolves to a unique passage and no title entered spoken or teleprompter text; use [outline scenarios](references/outline-scenarios.md) when validating changes
   - Anti-patterns: checking only that a file was written, leaving production instructions in generated-presenter copy
6. Hand off the script:
   - link to the markdown file
   - state whether it is ready for manual recording or generated-presenter input
   - mention scroll padding only when present
   - Anti-patterns: claiming the later visual plan is complete, implying the editing pipeline depends on script-level cues

## Anti-Patterns

- Adding YAML frontmatter or a metadata header to the generated script.
- Embedding motion graphics, visual direction, transitions, captions, crops,
  or edit notes in spoken copy without an explicit request.
- Adding teleprompter padding to a script that will be sent to HeyGen or another
  generated-presenter service.
- Including transcription, paper-edit, rendering, or API steps in this skill.
- Creating a landing page, deck, or editing plan unless the user asks.

## Output

Return:

- a link to the clean spoken-script file;
- whether it is ready for manual recording or generated-presenter input;
- a one-line note about visible scroll padding only when padding was requested
  and added.