agentsclimarketplace

Subtitles rewriter

Skill ur-grue/autopunk-media-skills/skills/tv-documentary/post-production/subtitles-rewriter

Reformats and rewrites subtitle text for timing, readability, and on-screen clarity, following broadcast subtitle standards.From its SKILL.md

Install
npx -y skills add ur-grue/autopunk-media-skills --skill subtitles-rewriter

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 21 stars21 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

7.0 KB, ~1.6k tokens by cl100k_base, as published. Nobody here has run it

Subtitles Rewriter

What This Skill Does

Reformats and rewrites subtitle text for timing, readability, and on-screen clarity, following broadcast subtitle standards.

When To Use This Skill

  • Auto-generated subtitles contain transcript errors that need correcting before delivery
  • Subtitle timing needs reformatting for broadcast delivery (line length, reading speed, two-line breaks)
  • Subtitles were originally written for one platform and need reformatting for another (cinema, broadcast, streaming, DVD)
  • Foreign-language dialogue subtitles need to be adapted for clarity and reading speed without re-translating

What You Need To Provide

Required: The subtitle text to be rewritten; the context (documentary title, type of content — dialogue, narration, foreign language, on-screen text); the target platform (broadcast, streaming, cinema, DVD/home video) Optional: The original timecodes if reformatting is needed; the target maximum characters per line; the target maximum words per second (reading speed); whether this is SDH (subtitles for deaf and hard of hearing) or standard subtitles; the language

How the Assistant Approaches This

  1. Reviews the existing subtitle text for transcript accuracy, line breaks, and character count — identifies errors and over-long lines
  2. Reformats to the target platform's line length and reading speed standards (see Output Format for typical values)
  3. Corrects errors in auto-generated transcript (common patterns: homophones, proper nouns misread, non-standard accents misread, punctuation errors)
  4. For two-line subtitles, applies grammatical line-break rules — breaks at clause boundaries, not in the middle of phrases

Output Format

Corrected subtitle text in the format provided (SRT, WebVTT, or plain numbered blocks). Each subtitle block: number, timecode (preserved or adjusted if requested), two lines maximum, maximum 42 characters per line (broadcast standard), maximum 17 words per second reading speed. For plain text input without timecodes: numbered blocks only, with notes on any corrections made. Corrections are listed separately in a Changes Log at the end.

Platform standards applied by default unless otherwise specified:

  • Broadcast (UK/international): 42 chars/line, 2 lines max, ~160–180 wpm reading speed
  • Streaming (Netflix/Amazon standard): 42 chars/line, 2 lines max, 17 wpm max
  • Cinema (DCP): 32–40 chars/line, 2 lines max, conservative reading speed

Quality Criteria

  • No subtitle block exceeds the platform's character-per-line limit
  • Line breaks occur at grammatical boundaries — not mid-phrase
  • Reading speed is within platform standards — no subtitle requires faster-than-standard reading
  • Proper nouns, technical terms, and names are corrected where auto-generation has produced errors
  • Changes Log accurately records every correction made

Example

Input

Platform: Broadcast (UK standard) Content: Documentary interview — foreign language dialogue (French), English subtitles Context: Interview with a marine biologist discussing ocean pollution

Auto-generated subtitles (with errors):

1
00:02:14,400 --> 00:02:18,900
the microplastics they are entering the food chain at every level but we don't see them

2
00:02:19,200 --> 00:02:24,800
the fish consume them and then we consume the fish and nobody is talking about the concentration levels that we are now measuring in human blood samples

3
00:02:25,100 --> 00:02:27,300
it's not a projection anymore it's already happening

4
00:02:31,000 --> 00:02:37,200
in 1972 the scientist jock cousteau already described the Mediterranean as a garbage dump nobody listened

5
00:02:37,500 --> 00:02:42,100
and now here we are fifty years later and the situation is categorically worse not better

Output

1
00:02:14,400 --> 00:02:18,900
Microplastics are entering the food chain
at every level. We don't see them.

2
00:02:19,200 --> 00:02:22,600
The fish consume them,
and then we consume the fish.

3
00:02:22,700 --> 00:02:24,800
Nobody is talking about the concentration levels
we're now measuring in human blood samples.

4
00:02:25,100 --> 00:02:27,300
This isn't a projection anymore.
It's already happening.

5
00:02:31,000 --> 00:02:34,800
In 1972, the scientist Jacques Cousteau
described the Mediterranean as a garbage dump.

6
00:02:34,900 --> 00:02:37,200
Nobody listened.

7
00:02:37,500 --> 00:02:42,100
Fifty years later, the situation
is categorically worse. Not better.

Changes Log

BlockChangeReason
1Added capitalisation; restructured from run-onBroadcast standard; readability
2Split from original block 2; shortened each lineOriginal block 2 was over 180 words per minute at this timecode; split into two subtitles with adjusted timecodes
3New block from end of original block 2Overflow from split above
4"it's not a projection anymore" → "This isn't a projection anymore"Cleaner register for subtitle reading; subject pronoun added for clarity
5"jock cousteau" → "Jacques Cousteau"Proper noun correction — auto-generation phonetic error
5Split into two blocksOriginal block 5 was 22+ seconds with ~190 wpm; split required
6"Nobody listened" separated as standalone blockCreates emphasis; fits naturally after the Cousteau subtitle
7"here we are fifty years later" → "Fifty years later"Reduced characters; removes register-inappropriate colloquial phrase for subtitle context

Known Limitations

  • This skill corrects subtitle text based on the input provided. If the underlying transcript is wrong (mis-heard speech, incorrect translation), the skill corrects formatting errors but cannot know the content error. Provide accurate transcript text as input.
  • Timecode adjustment (when a subtitle block is split or merged) is provided as a best estimate based on the text length. Precise timecodes must be confirmed against the actual audio in a subtitle editing application.
  • Subtitle standards vary by broadcaster, territory, and platform. The defaults used here are broadcast-standard UK/international. Different platforms (Netflix, Amazon, Disney+) have specific style guides that may differ in small but delivery-critical ways — check the commissioning platform's technical specification before final delivery.
  • SDH subtitles (including sound effects, music descriptions, and speaker identification) have different requirements that are not fully addressed by this skill's standard output. Flag SDH content explicitly when requesting a rewrite.

Related Skills

What ships with it: 1 file

3.1 KB alongside SKILL.md

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.