Plugin architecture

[theophile-scrive]

We want to enable people to write their own plugins for Kitab. As such, a flexible but structured mechanism is needed. Today, I'm thinking that the following things would be good:

• A plugin is an executable named kitab-<plugin-name>, invoked by passing --plugin=<plugin-name> to the CLI.
• Kitab will spawn that executable
  • (optional?) Have the plugin "register" itself (purpose, position in time pipeline?). To be refined.
  • send the graph and inventory data in JSON on stdin.

Plugin registration

It could be interesting, for transparency and debugging, that plugins declare what they intend to do. This way, it can be outputted by Kitab as part of the verbose logging and there is no bad surprise.