I've been thinking a lot about what Agent Canvas unlocks--especially wrt Automations. Automations are the biggest value add on top of existing agents like Claude Code The main things folks seem to love about Hermes and OpenClaw is the ability to trigger them from e.g. WhatsApp or Slack. This is exactly what automations does well! So for our initial launch, we should highlight the ability to build a GitHub or Slack resolver. We'll have an easy way to connect those via an API key, and then set up an automation which polls a channel/repo for events, and responds appropriately. This will probably involve shipping a built-in skill that tells OpenHands how to build these automations well

Sounds great For setups like github or slack do we want the user to go through a conversation flow, or offer presets which they can toggle immediately?

it's a great question, but robert, i, and a few others discussed and we thought that going through a conversation flow might actually make people think a little bit more about what is possible. it highlights the flexibility.

Would it be possible to potentially enable a "one-click install" of these resolvers within Agent Canvas? I think Automations is able to execute scripts built around the SDK (I think). If so, we could theoretically have a python script that handles the resolver workflow (e.g., for Slack, acknowledging receipt of the

@openhands

request, commenting on thread with the final message). Having a Skill that helps create the Automation is obviously another path. But wondering if there's an easier, less error-prone initial starting point as well that comes native.

Would it be possible to potentially enable a "one-click install" of these resolvers within Agent Canvas?

I had a similar idea a while ago about having a "Add backend with Agent" feature 🙂 I think that's 100% do able, we just need to come up with a prompt that starts the convo, giving the agent details like: 1. you are assisting a user to setup "XX" 2. interview the user for all the configuration choice, explain to the user what does each option means, give them your recommendation 3. then help user setup automation using the "openhands-automation" skill https://github.com/OpenHands/agent-canvas/issues/233

The only issue with a "one click install" for these bots is that you need to specify certain things e.g. for a slack responder, we can't poll every channel--you'll need to tell us what channel you want or for GitHub, you might want to restrict to a particular repo

i think after you click "install", it will take you to a new convo where the agent will ask you questions

Right. I think that's mainly because the polling approach we need to take is going to create limitations for the user (e.g., you need to specify channels or specific repos; since it's not event-driven, the user will need to make tradeoffs). But I also think that might help make the Event-driven Integration Hub a more obvious "next step" for users.