CLI Agent Workflow Evaluation Bench

Workflow Profile

READY

Target Task Pipeline

Codebase Scale 85 kLOC

Test Run Frequency Every Step

Execution Guardrails Bash Confirmation

Context Budget 128k Tokens

Claude Code (CLI Agent) Codex (Inline Assist)

claude-code (sub-agent loop) 0.0s

codex / copilot (single-file) 0.0s

2.8x EFFICIENCY

Efficiency Multiplier

2.8x

vs Manual Copy-Paste

Dev Context Switches

1 vs 8

Agent vs Codex

Self-Correction Cycles

2 auto

compiler test loop

Token Overhead

-42%

Grep vs Entire File Dump

Terminal Agent: 75% Inline Assistant: 25%