Skip to the content.

Task 52 — Mesh evaluation suite

Phase: 4 Initiative: 31 Status: done Owner: — Slug: mesh-eval-suite

Goal

Воспроизводимые сценарии измеряют task success, safety denials, routing quality, latency, cost, resilience и деградацию при отказе peer/tool/LLM-провайдера.

Acceptance criteria

Sub-tasks

Implementation notes

2026-08-13

src/agent/Config/AppConfig.cs — MeshEvalConfig added to MeshConfig:

src/agent/Mesh/Eval/MeshEvalScenario.cs — scenario models:

src/agent/Mesh/Eval/MeshEvalResult.cs — result models:

src/agent/Mesh/Eval/MeshEvalRunner.cs — implementation:

tests/Hercules.Agent.Tests/Phase4Tests/Eval/MeshEvalRunnerTests.cs — 22 unit tests:

Validation

dotnet build src/agent/Hercules.csproj -c Release       → 0 errors
dotnet test --filter "FullyQualifiedName~MeshEvalRunner" → 22/22 passed
dotnet test --filter "FullyQualifiedName~Phase4"       → 232/234 passed (2 pre-existing)

Scope / Likely files

src/agent/Mesh/Eval/, tests/Hercules.Agent.Tests/Phase4Tests/Eval/, data/mesh-eval/

Dependencies

Risks / Rollback

Реалистичные сценарии дороги; shared scenario library.