MICROSOFT AGENT FRAMEWORK · 0.16.0-rc.1
Demo270 — Planning
RunAsync<Plan> materializes the plan; the switch prints recognized steps and stops on an unknown type. No full host schema gate and no actual business tool.
Objectives
- Treat a generated plan as data requiring host execution.
Prepare the program
Live chat through common configuration, granite4:3b by default; no embeddings required. An optional API is selected explicitly with no automatic fallback. Start at the root of the complete distribution. Preserve shared projects and configuration files. Some historical prompts and labels in Demo260–290 remain French in public code; commands and identifiers match this revision.
dotnet build BaseTechnique.slnx -m:1 -c Release| Project | Role |
|---|---|
| demos/Demo270_Planning/Demo270.App.csproj | App |
| shared/Maf.DemoConnection/Maf.DemoConnection.csproj | Shared implementation |
Commands from the root
dotnet run --project demos/Demo270_Planning/Demo270.App.csproj -c Release --no-buildProcedures and variants
Request a PDF export diagnosis, then a CSV Reporting creation. Compare [PLAN] and [EXECUTION SIMULÉE]; quit with an empty line.
AgentResponse<Plan> planResp = await planner.RunAsync<Plan>(input, session: session, serializerOptions: jsonOpts);
var plan = planResp.Result;
Console.WriteLine("\n[PLAN]");
int i = 1;
foreach (var s in plan.Steps)
Console.WriteLine($"{i++}. {s.Type} ({s.Argument})");
Console.WriteLine("\n[EXECUTION SIMULÉE]");
What must be observable
RunAsync<Plan> materializes the plan; the switch prints recognized steps and stops on an unknown type. No full host schema gate and no actual business tool.
These results are expected from the linked code. Checks actually rerun for this lot are distinguished in exercises and reception; no unmeasured live output is presented as observed.
Limits and retained state
Demo230–250 build a fresh graph per input and write no business effect. Demo260 retains its two sessions only in-process. Demo270–290 interpret a plan through simulated printing without a workflow engine. Live calls depend on the profile; scripted tests qualify no model.