Video Content Renamer

• Treat locations, addresses, and building names as hypotheses unless the video visibly contains exact text or the user already supplied the location.
• Use direct visual descriptions when confidence is limited:
  • kitchen_person_cooking
  • night_city_lights
  • street_interview_two_people
• If a landmark is visually distinctive, it is acceptable to name the landmark, but avoid inventing exact addresses.
• If multiple labels are plausible, choose the least committal one and tell the user it is a visual inference.

Naming Rules

• Keep the rename pattern stable: 001_scene_subject_originalstem.ext.
• Preserve the original stem at the end of the new filename for traceability.
• Match the user's language when naming files. Chinese labels are fine if the folder already uses Chinese names or the user is working in Chinese.
• Keep labels short and searchable. Prefer 2-4 semantic chunks instead of full sentences.
• Avoid punctuation that is awkward in filenames. Use underscores as separators.

Review Asset Generation

Run:

python3 scripts/build_review_assets.py "/path/to/videos"

Important outputs:

• index.csv: one row per clip with duration, dimensions, and thumbnail path
• sheet_01.png, sheet_02.png, ...: contact sheets for multimodal review
• thumbs/: extracted representative frames

Use --output-dir when the source folder should stay clean or when the folder is large.

Rename Plan Creation

Create a CSV with at least these columns:

• old_name
• new_name

Chinese headers 原文件名 and 新文件名 also work.

Recommended process:

1. Review sheets.
2. Draft labels.
3. Save the plan as 视频重命名对照表.csv or video_rename_plan.csv.
4. Preview the rename operation before applying it.

Applying Renames Safely

Preview first:

python3 scripts/apply_rename_csv.py "/path/to/视频重命名对照表.csv"

Apply after verification:

python3 scripts/apply_rename_csv.py "/path/to/视频重命名对照表.csv" --apply

The script validates collisions and uses a two-phase rename so swaps and overlapping targets remain safe.

Practical Notes

• Prefer using a frame around 25% into the clip; very short videos should use the midpoint.
• For interview footage, name the visible setup first, not the inferred topic.
• For cooking, product, architecture, or travel footage, separate wide shots from close-ups in the label.
• Tell the user when a rename includes interpretation rather than plainly visible facts.

Open-Source Repo Notes

• When this skill lives in a Git repository instead of ~/.codex/skills, run the scripts from the repo root with relative paths as shown above.
• To install it into Codex manually, copy the repository folder into ${CODEX_HOME:-$HOME/.codex}/skills/video-content-renamer.