I’m sure @agiletortoise has good reasons for storing actions in a database that you can only edit through the UI. It likely prevents a lot of user error, keeps things performant and consistent, and makes backward compatibility easier to manage.
That said, as I’ve started building more complex actions—and especially as I “vibe code” small tweaks whenever I get a new idea—I keep wishing actions were plain text files organized in a normal folder structure. Being able to edit them directly, diff them, version them, and move them around like code would make iteration so much faster.
We’re also entering an era where more people will build tiny, personal automations to solve their own pain points. Drafts is already one of the best apps for this: it has a strong framework for user-created solutions. Historically, the limiting factor was the time and programming knowledge required. LLM-assisted coding lowers that barrier a lot, which means many more people can create and refine custom actions tailored to their workflows.
I know you’re already exploring how LLM integration can help inside actions. My argument is that the next step is using LLMs to help people create and maintain actions in the first place—especially if those actions can live as readable, editable text outside the app.
Of course, there are real risks here too: an easier pipeline could flood the action library with low-quality “slop,” and it could make it easier for people to run unsafe or careless code. But I still think there’s something worth exploring in a more code-native, file-based workflow—ideally with guardrails.