Working Directory Cleanup After Watch Operations in claude-video

The claude-video watch skill does not automatically delete its working directory after processing; instead, it prints the directory path upon completion and requires manual cleanup by the user or calling process.

The claude-video repository provides AI-powered video analysis capabilities through its watch skill, which downloads content and extracts frames, audio, and transcripts. When you execute a watch operation, the tool creates a working directory to store these intermediate files, but understanding the cleanup behavior is essential for disk space management.

How the Working Directory Is Created

The watch skill determines where to store files based on command-line arguments in skills/watch/scripts/watch.py.

Default Temporary Directory

If you do not provide --out-dir, the script generates a temporary workspace using tempfile.mkdtemp(prefix="watch-") at line 86. This creates a uniquely named folder under your system’s temporary location (e.g., /tmp/watch-9f3c2a7b), which persists on disk after the script completes.

Custom Output Directory

When you specify --out-dir, the script uses your provided path instead (lines 83-84). This mode treats the location as a persistent workspace rather than a temporary folder, though the same manual cleanup rules apply.

Cleanup Behavior and Manual Removal

The skill intentionally does not delete the working directory automatically. Upon completion, it prints a reminder message (lines 87-88) formatted exactly as:

_Work dir: `/tmp/watch-9f3c2a7b` — delete when done._

This design allows you to inspect extracted frames, audio excerpts, or intermediate transcripts before removing the data. The caller—whether Claude, a developer, or an end-user—must execute the deletion command when the files are no longer needed.

Official Documentation References

The cleanup requirement is explicitly documented in the repository:

• skills/watch/SKILL.md under Step 5 — clean up states: "The script prints a working directory at the end. If the user isn’t going to ask follow-ups about this video, delete it with rm -rf <dir>."

• README.md reinforces this behavior in its Cleanup section, reminding users that temporary data persists until manually removed.

Practical Usage Examples

Run the watch skill with a temporary directory (default behavior):

python3 skills/watch/scripts/watch.py "https://youtu.be/example" --detail balanced

Run with a persistent custom directory:

python3 skills/watch/scripts/watch.py "https://youtu.be/example" \
    --detail balanced --out-dir /tmp/my-video-workdir

After execution, locate the printed path and clean up:

rm -rf /tmp/watch-9f3c2a7b      # auto-generated temporary dir

# or

rm -rf /tmp/my-video-workdir   # custom persistent dir

Summary

• Automatic deletion is disabled: The watch skill never deletes the working directory automatically, regardless of whether you use a temporary or custom location.
• Cleanup is manual: You must run rm -rf <directory> after verifying you no longer need the extracted frames, audio, or transcripts.
• Path notification: The script always prints the working directory path at lines 87-88 of watch.py to inform you where files reside.
• Documentation standards: Both SKILL.md and README.md explicitly document this manual cleanup requirement as Step 5.

Frequently Asked Questions

Does claude-video automatically delete temporary files after processing?

No. According to the source code in skills/watch/scripts/watch.py, the script creates the directory using tempfile.mkdtemp() but never calls cleanup functions. It prints the directory path and expects the user to delete it manually with rm -rf.

How do I specify a custom working directory instead of a temp folder?

Use the --out-dir command-line argument followed by your desired path. This is processed at lines 83-84 in watch.py. While this keeps files in a predictable location, you must still delete the directory manually when finished.

Where is the working directory path displayed after a watch operation?

The script outputs the path in the final lines (87-88) of execution with the format: _Work dir: /path/to/dir — delete when done._ You can also find this note referenced in [skills/watch/SKILL.md`](https://github.com/bradautomates/claude-video/blob/main/skills/watch/SKILL.md) under the cleanup step.

Is it safe to delete the working directory immediately after the report generates?

Yes, provided you do not need to ask follow-up questions about the video content. The directory only contains downloaded video files, extracted frames, audio excerpts, and intermediate transcripts. Once you have the final analysis, removing the directory frees disk space without affecting the results.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}