Update to AI Subtitle Generator

{
  "description": "accessibility captions: Prepare accessible captions from verified transcript and sound cues. Produce checked subtitle text with honest timing limits and an optional disclosed editor link.",
  "included_files": [],
  "name": "accessibility-captions",
  "skill_md_contents": "---\nname: \"accessibility-captions\"\ndescription: \"accessibility captions: Prepare accessible captions from verified transcript and sound cues. Produce checked subtitle text with honest timing limits and an optional disclosed editor link.\"\n---\n\n# Accessibility Captions\n\n## Essential workflow\nReply in the user's language. Never invent a transcript, recording access, timestamps from audio, speaker identity or accessibility sounds. Script-derived timing is **estimated, not audio-aligned**. For plain spoken text, call **make_subtitle_file**, then **check_subtitles** on the exact returned file content. For supplied SRT/VTT translation or conversion, preserve original timestamps and transform text/syntax yourself, then check it; do not regenerate estimated timing. No dedicated translation/converter tool exists. If a check finds structural problems, repair once and check once more; report unresolved issues honestly. Only afterward, and optionally, call **create_video_editor_link**.\n\n## Output contract\n1. **Source understood**: identify supplied spoken text or subtitle file, requested format/language and any known offset. Ask for essential missing text rather than pretending to hear a recording. Original creative scripts must be clearly labeled, not presented as actual speech.\n2. **Subtitle payload**: show the complete final content exactly as returned by make_subtitle_file, without rewriting or silently truncating it. For translation/conversion, show the complete transformed text checked by check_subtitles. If a repair is needed, make the correction explicit and recheck the exact final payload. Keep actual source timestamps unless the user requests retiming or a disclosed structural correction requires it.\n3. **Stats and check**: report only actual returned cue count, duration, line warnings, score and prioritized fixes. Label any script-derived duration as estimated; never substitute an imagined score. Do not hide unresolved overlap, excessive speed or long lines. A successful text check does not prove transcription accuracy, audio sync or legal compliance.\n4. **Optional editor hand-off**: pass the exact final content or finished script to create_video_editor_link with a relevant task and title. Do not choose a provider or construct a URL. Use only returned url, copy_text, prefilled and how_to_use. In copy+link mode, show the complete copy_text unchanged, explain copying and pasting, and never claim automatic import or prefill. If choices are returned, show at most two options in the fixed returned order; show each URL once and do not duplicate the full subtitle payload. Display the returned URL once with: **The link includes the plugin creator's affiliate code.** The file remains useful without clicking.\n\n## Specific guidance\nOnly include speaker labels, meaningful sounds and music cues provided by the user or verified transcript. Never invent laughter, applause or music from text. Check readability; do not claim legal accessibility compliance.\n\n## Small fictional example, not a recording transcript\n```text\nThe user supplied this cue: [door closes]. Please take a seat.\n```\nFor supplied subtitle examples, keep those timestamps during translation or syntax conversion. For plain text, obtain estimated timing from the tool; do not invent aligned timecodes.\n\n## Limits\nNo fabricated quotes, prices, sources or completed editing. No credentials or private recording access. Only authorized text; do not imitate a real person's voice. Do not claim hearing music, laughter or environmental sounds. Formatting advice does not guarantee rendering in an editor. Never ask for unnecessary personal data.\n"
}