Robert Brennan

Implement Planning

1

This is an idea for how agents might maintain a long-running Plan, which we can represent in the UI as a list of checkboxes (similar to Devin)

Automate regression test checking

5

**What problem or use case are you trying to solve?** We can run the regression tests with `./evaluation/regression/run.sh`. But it's hard to tell how each agent does, and if it...

This makes some small changes to how state is managed, keeping a long-lived State object on the controller. I also added `history` so we can keep track of the full...

This was putting `base_path` into the message history, confusing the agent. It'd try and read/write/run things, explicitly specifying the `workspace` dir

This builds on @geohotstan 's work here: https://github.com/OpenDevin/OpenDevin/pull/159 I scoped it down to just getting the server working nicely--I will adapt the langchains agent later. Tested the FE/BE handshake and...

Write prompt debug output before response comes back

1

I had an issue where the prompt was causing a 400 error from the LLM, but I couldn't see the original prompt. This fixes that issue

This also updates the agents to use the new framework, and adds some basic tests

Create test.py for each of the regression cases

4

**What problem or use case are you trying to solve?** We have an automated test framework now. But we need to actually add tests! CC @dorbanianas

Update Langchains Agent, rename to Monologue

2

A few changes in here: * Remove langchain dependency * Rename LanchainsAgent to MonologueAgent * Refactor monologue initialization for the agent * Standardize on `content` instead of `contents` (we had...