Agent Claude Code
Impartial text-only evaluator for the mcp-eval skill. Compares two code-research answers to the same task and produces a structured JSON verdict. Has no tools — judges from text alone to avoid evidence-based bias.
Agent Claude Code
Read-only code-research sub-agent restricted to the code-index MCP. Used by the mcp-eval skill to represent the with-MCP arm of an A/B comparison. Do not invoke directly outside that skill.
Agent Claude Code
Baseline read-only code-research sub-agent restricted to default file tools (Read, Grep, Glob, LS). Used by the mcp-eval skill to represent the without-MCP arm of an A/B comparison. Do not invoke directly outside that skill.