AI Agent Modules: Use memory (otherwise, task restarts when workspace restarts)

[bpmct]

Agents like Claude code Support memory (https://docs.anthropic.com/en/docs/claude-code/memory). When memory is enabled on the module.

When memory is enabled on the module, the system prompt and the module itself should keep track of the steps it took to complete a task. That way, when the user restarts their workspace, the agent will understand where it is at with the task instead of starting from the beginning. This should obviously be a local memory. I think Claude recommends doing this via an import, so let's do it that way.

It would also be neat if the agents could support committing to the project memory as it learns new things about the project.

Suggestion

module "claude-code" {
  task_memory = true
  update_project_memory = true
} 

Let's definitely start with Claude Code. I'm not sure to what extent Goose and Aider do this.

[hugodutka]

The newest version of the Claude Code module already supports continuing sessions between workspace restarts. Goose doesn't - I tried making it work but ran into some bugs in Goose that made it too difficult in the time I had. Each agent has a slightly different way of configuring memory, so we should probably open a separate issue for every agent we'd like to support.