在 Manus 中运行任何 Skill
一键导入
一键导入
一键在 Manus 中运行任何 Skill
开始使用ab-compare
Generate and compare multiple implementation variants with benchmarking and scoring
星标7
分支5
更新时间2026年6月3日 02:58
文件资源管理器
2 个文件SKILL.md
readonly来源
安装命令
适用职业SOC
菜单
Generate and compare multiple implementation variants with benchmarking and scoring
Advanced browser automation for AI agents with snapshot-ref interaction pattern - navigate, snapshot interactive elements with refs, click/fill/select by refs, manage sessions, and extract structured data
Design RESTful APIs with best practices, conventions, and OpenAPI specs
Generate API documentation from code - produce OpenAPI/Swagger specs, Markdown API references, request/response examples, and interactive documentation from source code analysis
Universal API gateway - connect to 100+ APIs with a unified interface, manage API keys, and chain API calls
API测试技能 - IPC处理器发现、测试生成、健康检查、回归测试
双阶段架构模式 - 先规划后编辑，分离推理与代码修改，提升多文件编辑准确率
| name | ab-compare |
| display-name | A/B Compare |
| description | Generate and compare multiple implementation variants with benchmarking and scoring |
| version | 1.0.0 |
| category | development |
| user-invocable | true |
| tags | ["ab-test","compare","variants","benchmark","code-generation","multi-agent"] |
| capabilities | ["variant-generation","benchmarking","scoring","decision-recording"] |
| tools | ["agent_coordinator","code_reviewer"] |
| instructions | Use this skill when the user wants to compare multiple implementation approaches. Generates N variants (default 3) using different agent profiles (concise, robust, readable, performant, testable), benchmarks each on error handling, readability, and conciseness, then ranks and selects a winner. Usage: /ab-compare "task description" [--variants 3] [--benchmark] |
| examples | [{"input":"/ab-compare \"add input validation to login form\"","output":"3 variants generated. Winner: robust-agent (score: 78/100)"},{"input":"/ab-compare \"implement retry with exponential backoff\" --variants 4","output":"4 variants compared. Winner: performant-agent (score: 82/100)"}] |
| os | ["win32","darwin","linux"] |
| author | ChainlessChain |
| handler | ./handler.js |
Generates multiple implementation variants for a task using different agent profiles, benchmarks each variant, and selects the best one.
/ab-compare "task description" [--variants 3] [--benchmark]
| Profile | Strategy |
|---|---|
| concise-agent | Minimal code, one-liners, built-in methods |
| robust-agent | Comprehensive error handling, edge cases |
| readable-agent | Self-documenting, clarity over brevity |
| performant-agent | Optimized, minimal allocations |
| testable-agent | Pure functions, DI, clear interfaces |
| Criterion | Weight | Description |
|---|---|---|
| Conciseness | 30% | Fewer lines = higher score |
| Error Handling | 35% | try/catch, null checks, throw |
| Readability | 35% | Line length, comments, naming |
A/B Comparison Report
=====================
Task: "add input validation to login form"
Variants: 3
| Agent | Concise | Error | Readable | Total |
| -------------- | ------- | ----- | -------- | ----- |
| concise-agent | 90 | 50 | 60 | 65 |
| robust-agent | 40 | 95 | 70 | 78 |
| readable-agent | 60 | 65 | 90 | 73 |
Winner: robust-agent (78/100)