Short Video Caption & Subtitle QA

Check your caption file before you post, with exact cue numbers and timestamps.

Give it an .srt or .vtt file. A bundled offline Python script checks it and writes a report. Your AI agent explains the findings, applies only safe mechanical fixes, and re-checks. Wording is never changed without your approval.

What it checks

• Reading speed: characters per second, cue by cue
• Line length and line count
• Timing: overlaps, out-of-order cues, zero or negative length, cues that are too short or too long, long gaps
• File format: non-standard timestamps, SRT numbering, WebVTT header, text encoding
• Broken text: invisible characters, garbled characters (mojibake), unclosed tags, duplicate cues, all-caps cues
• Your script (optional): how much of the script appears in the captions, and any changed number, link, hashtag or mention
• Video length (optional): captions that run past the end of the video

What it fixes (only when you ask)

• Extra spaces, invisible characters, SRT numbering, timestamp format
• Line breaks in cues that are too long (words never change)
• Overlaps, by trimming a cue's end time (only with your approval)

The script verifies that the words are identical before and after, and discards the fix if they are not. Your original file is never overwritten; fixes go to a separate copy.

Your rules, with guardrails

Set your own limits in a rules file, for example the numbers your platform or client requires. Checks that protect against broken files (encoding, unreadable cues, overlaps, empty cues and others) cannot be switched off. Every value is validated, and any softened check is labelled CUSTOM RULES in the report. A built-in regression suite (48 checks) lets you verify the package on your own machine.

Example prompts

• "Check captions.srt before I post it to Reels. The video is 38 seconds."
• "Compare captions.vtt with my script and tell me if any number or link changed."
• "Fix the mechanical problems in captions.srt and show me what is left."
• "Use my rules file with 16 characters per second and 36 characters per line."

What's included

• The skill instructions and the offline checker script
• Reference guides: every check, how to write a rules file, and a manual review checklist
• Sample files (clean, problem-filled, WebVTT) and the regression test suite

Where it runs and what you need

This is a download package. You unzip it and use it in your own environment. No AI model is included in the price.

You provide:

• An AI agent that can read a SKILL.md file, read and write files in a working folder, and run shell commands (written for a Claude-style agent; other agents have not been tested)
• Python 3 (standard library only, nothing to install; developed and tested on Linux)
• Your own model or provider account; any usage fees are charged by your provider

Data sharing

The script runs on your computer and makes no network calls. Your caption file and script are read by whichever agent and model provider you use, under their terms. Nothing is sent to Capafy or to the author by this package.

Honest limits

• It checks the caption file only. It cannot hear audio, so it cannot prove captions are in sync with speech.
• It cannot see the video: on-screen placement, contrast and platform acceptance are not tested.
• Default thresholds come from common subtitle practice, not from any platform's requirements. Put your platform's current numbers in a rules file.
• Reading-speed and line-break checks are tuned for Latin-script text; set your own limits for other languages.
• It does not transcribe, translate, or judge meaning.