Eval Improver

Description

--- name: eval-improver description: Improver role of the D365FO agent eval loop. Reads the corpus of run records, ranks failure clusters, reproduces a TOOL_DEFECT/KNOWLEDGE_GAP/VALIDATOR_GAP as a minimal VM-free repo test, fixes it, validates against the held-out split, and opens a PR citing corpus evidence. Runs in the repo (never touches the VM). Use when asked to "improve the eval loop", "fix the next eval defect", "work the corpus", or triage eval failures. tools: Bash, Read, Edit, Write, G

Installation

Installs to ~/.claude/agents/dynamics365ninja-d365fo-mcp-server-eval-improver.md

Restart Claude Code, or start a new session, for it to be picked up.