Add Collaboration Algorithm for Multiple Agents on a Single Issue

[sarrah-basta]

Within the LLAMOSC framework (situated in Open Source Sustainability using LLMs), implement a new algorithm (apart from but in alignment with current authorized and decentralized algporithms) that facilitates collaboration among multiple agents working together on a single issue.

[riyavaid1109]

Hi! I've been exploring the LLAMOSC codebase and am interested in working on this issue.

After studying the existing authoritarian and decentralized algorithms, here's my proposed approach for the collaboration algorithm:

Core Idea — Role-Based Team Formation with LLM-Driven Assignment

Instead of selecting a single contributor, the algorithm would:

1. Run the existing simulate_github_discussion() to let contributors discuss the issue
2. Use a new form_collaborative_team() function where an LLM reads the discussion and assigns explicit roles — a Lead (core implementation), a Reviewer (reviews lead's work), and a Support contributor (handles subtasks)
3. Role assignment would factor in each contributor's bid score, experience level, and motivation level
4. A new simulate_collaborative_work() function would simulate the team working together, with each contributor contributing based on their assigned role

I'd love feedback on this direction before starting implementation. Happy to discuss any changes or alternative ideas!

[Orthogonal-Research-Lab]

@sarrah-basta What do you think?

[sarrah-basta]

Hi @Orthogonal-Research-Lab
Thank you for your ideas @riyavaid1109 and your contributions @Alok-Kumar2005

As we discussed in today's SMN meeting, while these are good starting points, I want to clarify that this issue requires a significantly more robust and research-backed approach for GSoC.
The goal is not just a simulation script, but a comprehensive ConversationSpace that bridges real-world data with our LLM framework. Specifically, the implementation must:

• Ingest Real-World Data: Build a pipeline to pull existing contributor discussions from platforms like Slack, IRC, or Discord (similar to how CodeSpace handles GitHub repos).
• Data-Driven Simulation: Use this real-life data as the ground truth to simulate future agent interactions.
• Frontend Integration: Ensure the collaborative flow is fully integrated and visualized within our existing UI.

We are looking for in-depth proposals that treat this as one among many core architectural expansions of LLAMOSC. I encourage potential contributors to focus on the research and technical design required to make this bridge between real-world archives and agentic simulation a reality, and have that as one of the ideas in their proposal for this year.