Play 07
Multi-Agent Service
Supervisor agent routes to specialist agents with shared state and handoff protocol.
A supervisor agent receives requests, classifies intent, and delegates to specialist agents (research, coding, data analysis, etc.). Each agent has its own model config, tools, and memory. Dapr provides state management and pub/sub messaging. Loop prevention and max-iteration guards keep costs predictable.
Architecture Pattern
Agent-to-agent handoff, supervisor pattern, shared state
Azure Services
DevKit (.github Agentic OS)
- agent.md — root orchestrator with builder→reviewer→tuner handoffs
- 3 agents — Multi-Agent Builder (gpt-4o), Reviewer (gpt-4o-mini), Tuner (gpt-4o-mini)
- 3 skills — deploy (107 lines), evaluate (108 lines), tune (109 lines)
- 4 prompts — /deploy, /test, /review, /evaluate with agent routing
- .vscode/mcp.json — FrootAI MCP with Azure OpenAI key + envFile
TuneKit (AI Config)
- config/openai.json — per-agent model params
- config/guardrails.json — loop prevention, max iterations
- config/agents.json — agent roles, delegation rules, fallback chains
- evaluation/test-set.jsonl — handoff scenarios
Tuning Parameters
Machine evidence
Certified runtime lifecycle
Maturity is calculated from contiguous, content-bound evidence. Missing or expired evidence demotes automatically; catalog claims cannot promote a play.
Deterministic evidence runner
Status: loading. Runs deterministic bounded fixtures at the edge with zero model calls, Azure operations, or cloud mutation.
Enterprise gate
Checks both What-If and production admission using server-derived maturity. It never calls Azure or deploys resources.
Repo Intelligence
v1A no-clone, revision-pinned map for agents and humans. Observed evidence is separated from inferred flow so the output stays useful without pretending to be a full call graph.