What problem does it solve? Agents running in a Kubernetes namespace have no built-in way to know which sibling agents exist or what they can do. This Skill builds a current, cached view of every reachable A2A agent without requiring cluster API access, extra RBAC, or new infrastructure. ## Core Features & Use Cases - Env-var based discovery: Reads the <SERVICE_NAME>_SERVICE_HOST / _SERVICE_PORT variables Kubernetes injects at pod start to enumerate candidate peers, skipping the API server and the agent itself. - Agent-card probing: Probes each candidate's /.well-known/agent.json endpoint to confirm it is a real A2A agent and capture its name, description, and declared skills. - Memory caching with pruning: Writes one reference-type memory entry per confirmed peer, updates the MEMORY.md index, and deletes stale entries for peers that no longer exist. - Use Case: After a pod restart, an agent needs to delegate a task. Run this Skill to refresh the peer list, then use the cached cards to pick a collaborator and hand off work with the call-peer skill. ## Quick Start Ask the agent to discover peers and refresh its list of available A2A agents in the namespace.
