Although our current approach using the co-star prompting technique has worked in most cases, we still encounter some datasets where the current prompts cause the agents to remain in a loop without solving the problem.
If you would like to experiment with agent prompts, this is an easy first issue to tackle using a trial-and-error approach. There are new techniques in prompt engineering that are constantly being tested, so feel free to explore and experiment.
Steps to Address This Issue:
- Navigate to the agent_manager.py script.
- Test new prompt techniques for each agent.
- If you believe your prompts perform better than the current ones, push your results!
Note: We advise omitting the .pkl files generated during the testing phase when pushing to Github.
Although our current approach using the co-star prompting technique has worked in most cases, we still encounter some datasets where the current prompts cause the agents to remain in a loop without solving the problem.
If you would like to experiment with agent prompts, this is an easy first issue to tackle using a trial-and-error approach. There are new techniques in prompt engineering that are constantly being tested, so feel free to explore and experiment.
Steps to Address This Issue:
- Navigate to the agent_manager.py script.
- Test new prompt techniques for each agent.
- If you believe your prompts perform better than the current ones, push your results!
Note: We advise omitting the .pkl files generated during the testing phase when pushing to Github.