Harness detail

Claude Code

Interactive terminal coding agent from Anthropic with tools, project instructions, and session workflows. Catalog entry only until an official or reverse-stable headless eval contract is pinned.

Preview operational data. Rows carry origin seed, mock-eval, or runner. Mock-eval and frontend-eval runner rows are not public Index ranks.

Version: catalogPreview
Source: externalPreview
Status: catalog-onlyPreview
Origin mix: No runsPreview

Source: https://github.com/anthropics/claude-code

Catalog notes: Included as the natural peer to Grok CLI for terminal-agent comparisons. Evaluation stays catalog-only until a documented headless path is frozen. Capabilities: tools, sub-agents, headless, mcp, skills, sessions, hooks.

Per-run scores as stored. This page does not invent a mixed-origin mean.
Run	Task	Origin	State	Score	Duration
Catalog-only harness. No preview runs until an adapter scores a frozen suite.