shield_lock
JinDaGe
← Back to skills
extension
Category: OtherAPI key required

糖果果字幕校对skill

Use when proofreading Jianying, Whisper, or other video .srt subtitles while preserving timestamps, subtitle order, and per-block line counts.

personAuthor: user_e70e31ebhubcommunity
downloadDownload package

puffer1021-subtitle-proofreader

Use this skill when the user provides a Jianying, Whisper, or other video .srt subtitle file and wants safe proofreading without changing timestamps or subtitle structure.

Inputs

  • Required: .srt subtitle file
  • Optional: reference article for background understanding only
  • Optional: provider choice, supports qwen, deepseek, glm

Outputs

  • *.fixed.srt
  • *.merge-report.md

Safety boundary

  • Keep subtitle order unchanged
  • Keep timestamps unchanged
  • Keep per-block line counts unchanged
  • Reference article is background-only, not rewrite source
  • If provider output fails validation, do not write .fixed.srt

Workflow

  1. Parse the .srt file and flatten subtitle text lines into stable numbered rows.
  2. Run two independent proofreading passes in chunks.
  3. Validate that model output contains exactly the expected numbered rows.
  4. Conservatively merge A/B candidates, preferring agreement and avoiding punctuation-only rewrites.
  5. Rebuild the original SRT structure with corrected text only.
  6. Write *.fixed.srt and *.merge-report.md next to the input file.

Provider selection

  • Natural language is allowed, for example: “用 deepseek 跑这份字幕”
  • CLI also supports explicit selection with --provider
  • Optional model override with --model
  • Environment variables: DASHSCOPE_API_KEY for qwen, DEEPSEEK_API_KEY for deepseek, ZHIPUAI_API_KEY for glm.

CLI examples

tgg-subtitle-skill input.srt --provider qwen --reference article.md

tgg-subtitle-skill input.srt --provider glm --model glm-4-flash --reference article.md

tgg-subtitle-skill input.srt --provider deepseek --chunk-size 150

Setup if the CLI is unavailable

If tgg-subtitle-skill is not installed, install the public package:

python3 -m pip install git+https://github.com/puffer1021/tgg-subtitle-skill.git

Then rerun the CLI command. Do not proceed with a fake output unless the user explicitly asks for a dry run or test mode.

Package layout

  • SKILL.md: trigger, inputs, outputs, safety boundary
  • README.md: implementation and engineering notes
  • jianying_subtitle_proofread/: Python implementation
  • examples/: example input and output files
  • tests/fixtures/: test-only subtitle datasets
  • references/: optional terminology and workflow notes