Agent-to-Agent (A2A) control plane for Muffy1's hybrid Android ⇄ local workstation ⇄ cloud agent stack.
Agents (Operit, Jenny, Termux-MCP, Kortix, Notte, OpenTerminal) advertise capabilities via agent cards and discover each other through a YAML registry, so they can hand off work instead of talking N×N.
a2a/agent-card.example.json— schema for describing an agent.a2a/registry.yaml— discovery registry (agents + health endpoints).a2a/README.md— how to register agents and collaborate.AGENTS.md— operating rules for humans + agents..github/workflows/agent-health.yml— scheduled health checks + validation.
- Copy
a2a/agent-card.example.json→a2a/agents/<name>.json, fill it in. - Add the agent to
a2a/registry.yaml. - Wire a self-hosted runner (or public endpoints) for the
agent-healthworkflow to ping local health URLs.
Local-first, private, self-hosted. Scoped tokens only — never commit secrets.