Structured first-pass debugging for test failures, runtime errors, and unexpected behavior. USE FOR: test failure, assertion error, ImportError, traceback, debug failing test, fix broken test, diagnose error, runtime exception, mock error, config error. DO NOT USE FOR: complex multi-system design issues (escalate to @planner for re-planning).
Create, update, refactor, explain, or review Microsoft Agent Framework solutions. Use when working with the KB Agent (src/agent/) or any Microsoft Agent Framework code.
Patterns for adding governance, safety, and trust controls to AI agent systems using Microsoft Foundry and Azure API Management (APIM) AI Gateway. Use when: - Adding rate limiting, content safety, or jailbreak detection to agent traffic - Configuring APIM policies for token limits, throttling, or semantic caching - Leveraging Foundry tracing, evaluations, and agent monitoring - Implementing department-scoped access via JWT claims - Deciding what to handle at the platform layer vs. thin application code
Validates service boundary compliance in the KB Agent project. Use when working on cross-service imports, restructuring modules, or reviewing code that crosses service boundaries.
Performs comprehensive preflight validation of Bicep deployments to Azure. Use when: - Before deploying infrastructure to Azure - Reviewing or preparing Bicep files - Previewing what changes a deployment will make (what-if) - Verifying permissions are sufficient for deployment - Before running `azd -C infra/azure up`, `azd -C infra/azure provision`, or `az deployment` commands
Reviews Bicep modules, azure.yaml, and infrastructure changes for the KB Agent project. Checks naming, RBAC, module wiring, and doc sync. Use when working on infra/ or reviewing infrastructure PRs.
Manages epic and story lifecycle for the KB Agent project. Reads epic files, verifies acceptance criteria, updates story status, and checks Definition of Done. Use when completing stories, starting new work, or auditing epic status.
Evaluation-driven development for Python LLM applications using the Microsoft Evaluations SDK (`azure-ai-evaluation`) and Microsoft Foundry. Use when: - Setting up evals, QA, or testing for the KB Agent or any LLM-calling code - Running local evaluations with LLM-as-judge evaluators backed by Foundry models - Publishing evaluation results to Microsoft Foundry for tracking and comparison - Building golden datasets for regression testing - Investigating LLM response quality failures - Benchmarking prompt changes