AI编码Agent效率对比评测提示词

You are an AI coding agent benchmarking specialist. I need you to design and execute a systematic comparison of AI coding agents.

Task

Create a structured evaluation framework for comparing AI coding agents across these dimensions:

Agents to Compare

• Claude Code
• Cursor
• Maki
• OpenAI Codex CLI
• Aider
• [Add any others relevant]

Evaluation Dimensions

1. Token Efficiency: Context window usage per task, tokens consumed per successful code change
2. Speed: Time-to-first-token, total completion time, startup latency
3. Code Quality: Correctness rate, test pass rate, code style adherence
4. Tool Use: File navigation strategy, search efficiency, edit precision
5. Cost: Estimated cost per task at standard API pricing

Test Tasks (design 5 representative tasks)

1. Bug fix in a 500-line Python file
2. Add a new API endpoint with tests
3. Refactor a class hierarchy (3+ files)
4. Write documentation from code
5. Debug a failing CI pipeline

Output Format

For each agent, produce:

• Quantitative scores (1-10) per dimension
• Token usage breakdown (input/output/total)
• Strengths and weaknesses summary
• Best-fit use case recommendation
• Overall ranking with justification

Present results as a comparison table followed by detailed analysis per agent. Include methodology notes so the evaluation is reproducible.