Prime-Agent walkthrough by snimu (Sebastian Müller)

Get started Install the latest stable release on macOS or Linux: curl -fsSL | sh

@omouamoua I've been heavily using prim-agent width a depth of 3 for the past few days. Great results so far. I think the idea of the single tool python kernel is fascinating and is contributing to the success so far.

@omouamoua I really like your harness, but it also has some problems. The most important issue is that sometimes when your harness write a test and the test runs, it gets stuck and the process doesn’t continue. There is no command to interrupt the process and resume/continue.

You guys post total tokens but can you also post the input/output in prime agent? all that tooling in python the model has to write consumes output and even if prime agent uses half the token the total cost could still be higher that others. I have not benchmarked it myself but was just thinking if you guys have the breakup for it?

@omouamoua Nice - is this mostly showing tool-use and file edits, or also the planning/state UI? For an agent framework the hardest thing is making progress visible while it's working. What have been the weak spots so far?