feat: add /graph command to export conversation as Graphviz visualization

[modulovalue]

Summary

• Adds a new /graph slash command that exports the current conversation as a graph visualization
• Generates a Graphviz DOT file representing the conversation flow with user messages, assistant responses, and tool invocations
• Renders to SVG and opens in Finder for easy viewing

Features

• Color-coded nodes by type:
  • Blue: User messages
  • Yellow: Assistant text responses
  • Green: Read/Glob/Grep tools
  • Orange: Write/Edit tools
  • Purple: Bash commands
  • Light blue: Task agents
• Shows tool parameters (file paths, commands, patterns)
• Displays tool results with error highlighting
• Handles context compaction markers

Test plan

• Run vide and have a conversation with tool use
• Type /graph to export
• Verify SVG opens in Finder
• Check graph shows correct conversation flow
• Test error case when Graphviz is not installed

[modulovalue]

This is not a finished feature. This is just an idea that I think might be interesting.

@Norbert515 are there interesting test cases that we could use to test such a feature? I'm especially interested in seeing how, or if we can, visualize how the agents interact with each other.

I think such a visualization could be used to explore if there are any opportunities for optimizing context management.

Feel free to let me know what you think.

[dkbast]

@modulovalue this could be a great tool to investigate a refactoring process. I just used vide to update an old project of mine to the latest flutter version and bump dependencies. Since this might actually be a common workflow for using vide (both for hobby and "small" client projects) it could be worth building an agent for this case & optimising it using this visualisation.

[modulovalue]

this could be a great tool to investigate a refactoring process. I just used vide to update an old project [...]

That's like context management but for humans!

The overarching idea that I have is that I'd like to eventually play around with dynamic context, where we keep all the context around, but are able to provide summarized snapshots/views, depending on the task.

@dkbast Your use case is an interesting category where we could have a "refactored" view that only shows what has been changed, and if it's valuable to a human, it would most likely be valuable to agents as well.