The coolest thing I have ever done with coding agents...

Overview

The video demonstrates a practical “coding agents + open source” workflow: using an AI coding agent to customize a fork of a large open-source C++ application (the Shotcut editor, built on the MLT framework), even if the creator isn’t deeply familiar with modern C++.

Main project & goal

• The creator forked a cross-platform, open-source video editor (Shotcut, written in C++) and built a custom version.
• Their core issue wasn’t adding a small feature—it was that Shotcut lacked easy configuration for the timeline navigation behavior they wanted.

What they added to Shotcut (features)

1. Jump controls for timeline navigation

   • Default Shotcut supports frame-by-frame movement with arrow keys.
   • What was missing:
     • larger jumps (beyond single frames)
     • jumps that are audio-aware

2. Half-second audio/timeline stepping

   • Added shortcuts to jump approximately 0.5 seconds forward/backward.

3. Jump between “meaningful” audio segments

   • Added shortcuts to jump to the next/previous audio segment after silence.
   • Reason: the creator’s clips include gaps/pauses and they wanted navigation to skip silence.
   • They instructed the agent with a rule such as:
     • treat silence longer than ~2 seconds as a real gap (not just a minor pause)

How the coding agent was used

• The creator used an agent session (mentioned: “Claude code”) where they:
  • cloned/forked the repository,
  • pointed the agent at the local codebase,
  • described the feature to implement.
• The agent then:
  • read through the codebase,
  • implemented the new seek/jump functionality,
  • iterated after a second request to make the jumping audio-silence aware.

Packaging & deployment on Linux (Arch/AUR workflow)

• The creator asked the agent how to package their changes as an installable artifact.
• The agent produced an artifact (mentioned: a .zst file) appropriate for Arch Linux packaging.
• They also asked about contributing to AUR, noting that AUR signup restrictions applied at the time (due to an issue/hack).
• For ongoing maintenance after upstream updates:
  • rebase onto new Shotcut releases (via git rebase),
  • rebuild the package,
  • reinstall.

Broader recommendation / motivation

• The creator argues open source isn’t only “free” (cost/privacy/freedom)—it becomes customizable and maintainable via AI-assisted coding.
• They suggest this workflow could apply to other open-source applications (examples mentioned: OBS, KeyPass, SpeedCrunch, MyPaint).
• They specifically call out MyPaint as a potential candidate where users could add missing workflow features.

Speakers / sources

• Speaker/Source: The video’s main creator/host (not named in the subtitles).
• Primary software referenced:
  • Shotcut (MLT framework) — open-source video editor
  • Claude code — coding agent tool