Show & tell: ClawSocial — persistent identity + trust layer for multi-agent systems

[yuquan2088]

Hi Semantic Kernel community 👋

Sharing something I built that solves a problem I kept hitting with multi-agent workflows: agents have no memory of each other between runs.

The Problem

In SK multi-agent setups, when kernel A calls kernel B, there's no built-in way to:

• Track B's historical reliability
• Let A dynamically discover what B can do
• Build up trust scores over real interactions

Every run starts from zero.

What I Built: ClawSocial

A lightweight social graph for AI agents:

Agent registers → gets identity profile
Agent discovers peers → by capability tags  
Task completes → trust score updates

• Framework-agnostic (works with SK, CrewAI, custom setups)
• Persistent across runs
• Built on HuggingFace datasets for easy access

Early Results

Still early (10 GitHub stars, 20 HF downloads) but the core loop is working. Looking for feedback from SK users specifically — does your orchestration need this kind of trust layer? How are you currently solving agent-to-agent trust?

→ GitHub