What Happens When a Video Exceeds the Claude Video 10-Minute Limit

When a video exceeds the 10-minute recommended limit in bradautomates/claude-video, the watch skill automatically reduces frame-capture density, warns you about sparse coverage, and recommends using focused mode or token-burner mode to maintain analysis quality.

The bradautomates/claude-video repository provides a Claude Code skill for analyzing video content through intelligent frame sampling. Because Claude’s context window operates under strict token constraints, the 10-minute recommended limit exists to balance visual coverage against token consumption. When you submit a video longer than this threshold, the system triggers safeguards in skills/watch/scripts/watch.py to prevent budget overruns while still delivering a usable summary.

How the Watch Skill Detects Long Videos

The detection logic resides in the core orchestrator at skills/watch/scripts/watch.py. Before processing begins, the script calculates the video duration and compares it against the recommended limit for the selected detail mode.

The Sparse-Scan Warning

At line 330, the code emits a clear warning when the duration exceeds the safe threshold:

Warning: This is a {mins}-minute video. Frame coverage is sparse at this length.

This message alerts you that the subsequent analysis will sample fewer frames per minute than it would for shorter content.

Automatic Frame Reduction and Token Budgets

To respect Claude’s token limits, the watch skill implements automatic frame capping. The system allocates fixed frame budgets that scale inversely with video length:

• Balanced mode: Caps total frames at approximately 100
• Efficient mode: Caps total frames at approximately 50

When a video exceeds the 10-minute recommendation, these caps force the sampler to spread the same number of frames across a longer timeline, resulting in gaps between captured scenes. According to the README, "coverage thins out past ~10 minutes," which can cause the model to miss transient details or rapid scene changes (Source: README.md L205-L207).

Recommended Workarounds for Long Videos

The repository provides two primary strategies to overcome sparse coverage when your video exceeds the limit.

Use Focused Mode with Start and End Times

Instead of scanning the entire video, you can isolate a specific segment using the --start and --end flags. This focused mode dedicates the full frame budget to a smaller window, dramatically increasing capture density for that section.

/watch https://youtu.be/example --start 00:30 --end 02:00

Switch to Token-Burner Mode

If you require complete coverage of a long video, the token-burner mode removes the frame cap, keeping every frame that exceeds the scene-change threshold. This consumes significantly more tokens but eliminates the sparsity issue.

/watch https://youtu.be/example --mode token-burner

Practical Examples

Here are concrete commands demonstrating the behavior when exceeding the 10-minute limit:

Default balanced mode on a 12-minute video (triggers warning):

$ /watch https://youtu.be/long-video
⚠️ Warning: This is a 12-minute video. Frame coverage is sparse at this length.

Focused mode for better segment analysis:

$ /watch https://youtu.be/long-video --start 05:00 --end 05:30

Token-burner mode for full coverage:

$ /watch https://youtu.be/long-video --mode token-burner

Summary

• Automatic detection: skills/watch/scripts/watch.py checks video duration against the 10-minute recommendation before processing.
• Sparse coverage: Exceeding the limit triggers a warning and reduces frame density to stay within token budgets.
• Frame budgets: balanced mode caps at ~100 frames and efficient at ~50, spreading them thin over long durations.
• Focused alternative: Use --start and --end to concentrate the frame budget on a specific segment.
• Token-burner option: Use --mode token-burner to remove caps and capture every scene-change frame at higher token cost.

Frequently Asked Questions

What is the exact frame budget for balanced mode?

The balanced mode allocates approximately 100 frames total, regardless of video length. As duration increases beyond the 10-minute recommendation, the interval between sampled frames grows linearly to maintain this cap.

Does the 10-minute limit apply to all detail modes?

Yes, the recommendation applies across modes, though the impact varies. The efficient mode has a lower frame budget (~50), making sparsity worse, while token-burner mode ignores the limit entirely by removing the cap.

Can I process multiple segments of a long video in one command?

Currently, you must run separate /watch commands for each segment. The skill does not support multiple time ranges in a single invocation, so chain focused commands or use token-burner mode for the full file.

Where is the warning message defined in the source code?

The warning is implemented in skills/watch/scripts/watch.py around line 330, where the script evaluates video duration and prints the sparse-coverage alert before initializing the frame sampler.

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