## Fix Summary Refactored run_qa_agent() in workflow_v2.py to use label-based dynamic selection, similar to how triage was refactored in PLOW-7. ### Changes Made **1. Added new selector fields to StepConfig (config/loader.py:160-169)** - model_selector: Maps context values to model config keys (e.g., web → powerful) - mcp_selector: Maps context values to MCP server lists (e.g., web → [chrome-devtools]) **2. Added new methods to StepConfig (config/loader.py:271-367)** - resolve_model_key(): Resolves model config key from context - get_model_id(): Resolves actual Claude API model ID - resolve_mcp_servers(): Resolves MCP servers from context - resolve_all(): Resolves all dynamic values in one call **3. Created unified qa step (config/agents.yml:191-219)** - agent_selector: web→trckr-web-qa, cli→trckr-cli-qa, default→code-verification-qa - template_selector: web→qa-web.j2, cli→qa-cli.j2, default→qa-verification.j2 - model_selector: web→powerful (opus), cli/default→fast (sonnet) - mcp_selector: web→[chrome-devtools] **4. Refactored run_qa_agent() (workflow_v2.py:958-1018)** - Added _determine_qa_type() helper to map labels to qa_type - Simplified run_qa_agent() to use unified qa step with qa_type context - Reduced from ~105 lines to ~60 lines (including helper) **5. Updated _resolve_selector() and _build_options()** - Now passes resolved selectors for dynamic model/MCP selection ### Test Results - 436 tests passing - Added 30 new tests for new functionality - 73% code coverage maintained