Check subtitles before delivery

Subtitle Rescue audits basic SRT files and creates separate repair copies only when requested. Built for video editors, subtitle freelancers, course creators and adult production teams.

What it does

• Batch checks for missing or duplicate indices, invalid durations, out-of-order cues and nested overlaps.
• Configurable reading-speed, line-length, line-count and minimum-duration warnings.
• Markdown and JSON reports with source SHA-256 and before/after findings.
• Explicit repairs: normalize numbering, encoding and newlines; apply a fixed millisecond offset; sort chronologically; trim overlaps conservatively.
• Split strictly two-line cues into top and bottom tracks without guessing languages.

Example

Input: two bilingual cues numbered 7 and 7, at 00:00:01,000–00:00:03,000 and 00:00:02,800–00:00:05,000.
Request: normalize numbering, trim overlaps with a 50 ms gap, and export top/bottom tracks.
Output: cues numbered 1 and 2; the first ends at 00:00:02,750; the second is unchanged. Dialogue is preserved. Three initial warnings become zero findings under default checks. Outputs: repaired.srt, track-top.srt, track-bottom.srt, report.md, report.json and batch-summary.json.

Requirements and delivery

One-time Download, US$9.99. Includes readable Skill instructions, Python source, usage reference and license. Requires Python 3.10+ and an Agent Client with local file access and command execution. No third-party Python packages or publisher credentials. The standalone engine runs without a model; your chosen AI client and model charges are separate.

Limits

Basic SRT only. No transcription, translation, audio/video inspection, speaker detection or automatic synchronization. Maximum 100 files per batch, 5 MiB and 20,000 cues per file. Indices and timestamp hours: up to nine ASCII digits. Readability thresholds are heuristics. Intentional overlaps may remain. Invalid structure blocks subtitle exports. Check timing changes against source media.

Privacy

The local Python engine makes no network requests and collects no telemetry. Reports contain basenames, hashes and diagnostics, not dialogue. Your AI client may send subtitle content to its selected model provider according to its configuration and policies. This package does not choose or call a model provider directly.

Edition and support

English instructions and reports. A Traditional Chinese edition offers the same engine with localized documentation and Markdown findings; buy only the language edition you need. Subtitle text is never translated. JSON codes and technical errors remain English. AI explanations can be inaccurate; review the actual files. Support is available through the email on this listing. Refunds and usage rights follow the included Download License and applicable Capafy terms.