Cursor Forge

I use Hermes to think through work, gather context and decide what to do next. I also use Cursor for coding. Cursor Forge connects those parts of my workflow so handing a task to a coding agent doesn't mean starting over in another window.

The useful part is what happens after that first handoff. I want to know which run belongs to which task, what the worker changed, whether it finished, and what still needs attention.

From a plan to a coding task

Before a run, I narrow down the change. The worker needs to know which repository it is working in, what result I want, what it should leave alone and how we will check the result. I keep parallel coding work in separate workspaces so the agents don't edit the same checkout.

Cursor Forge gives me a handle for a run that I can refer to later. I can check its status, inspect the output and send a followup without losing the connection to the original task. That matters when a job takes longer than one conversation or needs a few rounds of review.

For example, I might ask a worker to fix a specific bug, then ask Hermes to inspect the change against the original report. If the test doesn't cover the failure, the next message can ask for that missing case instead of opening an unrelated task with half the context.

Getting the work back

A worker saying it finished is the start of review. I still want to see the diff, the tests and whether the change solves the problem I handed over. If it needs more work, the followup should be specific enough that the next attempt moves it forward.

That's what I'm trying to make easier with Cursor Forge: working with capable coding agents without becoming the person who spends the whole day moving prompts and summaries between them. I'm developing it for my own workflow, and the source is currently private.