port evaluation agent

[jtrayfield]

I ran it against two trajectory cases from earlier.

I had to fetch the characteristic form from hugging face.

[ShuxinLin]

did you run the refactored mcp and get the trajectories? I saw demonstration in trajectories which is completely removed in the new implementation. please don't add 0001.json and 0002.json. keep the repo clean. also add instructions in INSTRUCTIONS.md if needed.

@jtrayfield @DhavalRepo18

[jtrayfield]

where are the instructions for running MCP metaagent?

I removed 0001.json and 0002.json.

I added README.md

@ShuxinLin

[DhavalRepo18]

@ShuxinLin I suggest we address this GitHub issue together with the Evaluation Agent:
#205

We should introduce a base Trajectory class and then extend it for specific workflows (e.g., plan-execute). Other agents can similarly derive their own trajectory implementations. The Evaluation Agent should rely on this base class for defining evaluations.

Additionally, the instructions.md file is becoming quite large, let’s consider breaking it into smaller, modular, and independently manageable sections.

If possible, please add the base class as soon as you can and enable exporting trajectories from the plan-execute workflow so they can be consumed by the evaluation pipeline. It might also help to have a quick sync on your end to align on this.

Also, we should have a 1/2 example just to even run the code (without running the real plan/execute). Even if it is a dummy, it may help. But I will leave the decision up to you.