Skills DirectorySkills Directory
SkillsLearnSecurityCategoriesDocsBlogPro
Sign InSubmit Skill
Skills Directory

Security-tested agent skills for Claude, coding agents, and AI workflows.

Directory

  • Browse Skills
  • All Skills A–Z
  • Claude Skills
  • Claude Code Skills
  • Agent Skills
  • Categories
  • Authors
  • Submit a Skill

Learn

  • Learn Hub
  • Install Claude Skills
  • Write SKILL.md
  • Skills vs MCP
  • Directories Compared

Security

  • Security
  • Methodology
  • Secure Claude Skills
  • Security Badges
  • Chrome Extension
  • Skill Manager

Company

  • About
  • Community
  • Blog
  • API Docs
  • Advertise

2026 Skills Directory. All rights reserved.

ProTermsPrivacyRefunds
Back to skills

Coding Agent Headtohead Eval

ASecurity

当需要在自己代码库上把 Claude Code、Aider、Codex 等编码 Agent 做可复现的对比评测时使用;做基于 YAML 任务声明 + git worktree 隔离的多次运行,产出含通过率/成本/耗时/一致性的对比报告;不适用于评测纯对话/生成质量、单模型基准或无判据的开放任务。触发词:编码 Agent 对比、agent-eval、通过率、git worktree、选型评测

3 stars
0 votes
0 copies
2 views
Added 9/19/2026
ai-agentsgobashdockergitapi

Works with

claude codecursorcliapi

Security Analysis

A100/100

Scanned 9/19/2026

$npx -y skills add findscripter/everything-skills --skill coding-agent-headtohead-eval --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Coding Agent Headtohead Eval?

Add the live security badge to your README — it updates automatically with every re-scan.

Security grade badge for Coding Agent Headtohead Eval
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/findscripter-coding-agent-headtohead-eval/badge)](https://www.skillsdirectory.com/skills/findscripter-coding-agent-headtohead-eval)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
Files
SKILL.md
---
name: coding-agent-headtohead-eval
title: 编码 Agent 对比评测(通过率/成本/耗时)
description: 当需要在自己代码库上把 Claude Code、Aider、Codex 等编码 Agent 做可复现的对比评测时使用;做基于 YAML 任务声明 + git worktree 隔离的多次运行,产出含通过率/成本/耗时/一致性的对比报告;不适用于评测纯对话/生成质量、单模型基准或无判据的开放任务。触发词:编码 Agent 对比、agent-eval、通过率、git worktree、选型评测
domain: 智能/eval
triggers: [编码 Agent 对比, agent-eval, Claude Code vs Aider, Codex 对比, 通过率评测, pass rate, 成本对比, 耗时对比, 一致性 consistency, git worktree 隔离, YAML 任务定义, 编码工具选型, 模型更新回归检查, LLM-as-judge 判据, 数据化选 Agent]
tags: [评测, 编码agent, 对比评测, 通过率, 成本, 耗时, 一致性, git-worktree, yaml, 选型, cli]
level: 进阶
status: stable
agents: [claude-code, codex, cursor, aider, gemini-cli]
tools: [Read, Write, Edit, Bash, Grep, Glob, git, agent-eval]
requires: []
related: [llm-judge-evaluation, llm-agent-benchmarking, autonomous-coding-agent-patterns, ai-engineering-toolkit]
combines_with: [llm-judge-evaluation, parallel-agent-hub, llm-model-router]
license: MIT
source: affaan-m/ECC
source_license: MIT
---
# 编码 Agent 对比评测(通过率/成本/耗时)

用一个轻量 CLI,在可复现的任务上把多个编码 Agent 做「正面对决」(head-to-head)。"哪个编码 Agent 更好"通常靠感觉——本技能把它体系化为可度量、可复现的数据。

## 何时使用

适用:
- 在**自己的代码库**上横评 Claude Code、Aider、Codex 等编码 Agent,用数据而非感觉做选型。
- 采用新工具/模型前,先量化它在真实任务上的表现。
- Agent 更新了模型或工具链后,跑回归检查,看是否退化。
- 给团队提供有数据支撑的 Agent 选择依据。

不该用(负边界):
- 评测纯对话/文本生成质量、无代码改动的任务——用 `llm-judge-evaluation`。
- 只做单个智能体的能力/可靠性基准、对抗与回归门禁——用 `llm-agent-benchmarking`。
- 任务无客观成功判据(无法用测试/构建/模式匹配判定),且不愿引入 LLM 评审——评测无意义。
- 想修复或改进 Agent 本身的逻辑——转 `autonomous-coding-agent-patterns`。

核心理念:声明式任务 + 隔离运行 + 多次重复 = 可复现的对比,而非一次性的主观印象。

## 步骤

1. **声明任务**:建 `tasks/` 目录,每个任务一个 YAML(指定 repo、要改的文件、prompt、判据 judge、固定 commit)。
2. **跑对决**:对同一任务并列跑多个 Agent,每个至少 3 次(Agent 是非确定性的,需要看方差)。每次运行内部:
   1. 从指定 commit 创建全新 git worktree(隔离,免 Docker,互不干扰、不破坏基础库);
   2. 把 prompt 交给 Agent 执行;
   3. 运行 judge 判据;
   4. 记录 通过/失败、成本、耗时。
3. **出报告**:生成对比表,按通过率/成本/耗时/一致性横向比较,据此选型。

## 指令

CLI 用法(保留源命令):

```bash
# 1. 建任务目录
mkdir tasks
# 在 tasks/ 下按模板写每个任务的 YAML

# 2. 同一任务并列评多个 Agent,各跑 3 次
agent-eval run --task tasks/add-retry-logic.yaml \
  --agent claude-code --agent aider --runs 3

# 3. 生成对比报告
agent-eval report --format table
```

收集的指标(四维):

| 指标 | 测什么 |
|---|---|
| 通过率 Pass Rate | Agent 产出的代码能否通过 judge |
| 成本 Cost | 每任务 API 费用(可得时) |
| 耗时 Time | 完成的墙钟秒数 |
| 一致性 Consistency | 重复运行的通过率(如 3/3 = 100%) |

judge(判据)三类,按确定性优先选:
- **代码库 / 确定性**(首选):`type: pytest`(跑测试)、`type: command`(如 `npm run build`)。
- **模式匹配**:`type: grep`,用 `pattern` + `files`(如确认引入了 `class.*Retry`)。
- **模型评审 LLM-as-judge**(兜底):`type: llm` + 评判 `prompt`,用于难以用测试表达的语义判断。LLM judge 会引入噪声,**每个任务至少配 1 个确定性 judge**;LLM judge 细节见 `llm-judge-evaluation`。

最佳实践(保留源约束):
- **从 3-5 个任务起步**——选能代表真实工作负载的任务,别用玩具例子。
- **每个 Agent 至少跑 3 次**——非确定性,需要看分散度/一致性。
- **在任务 YAML 里固定 commit**——跨天/跨周结果才可复现。
- **每个任务至少 1 个确定性 judge**(测试/构建)——LLM judge 加噪声。
- **成本与通过率一起看**——10 倍成本换 95% 正确未必划算。
- **把任务定义纳入版本控制**——它们是测试夹具(fixture),当代码对待。

## 示例

任务 YAML(声明式,固定 commit 保可复现):

```yaml
name: add-retry-logic
description: Add exponential backoff retry to the HTTP client
repo: ./my-project
files:
  - src/http_client.py
prompt: |
  Add retry logic with exponential backoff to all HTTP requests.
  Max 3 retries. Initial delay 1s, max delay 30s.
judge:
  - type: pytest
    command: pytest tests/test_http_client.py -v
  - type: grep
    pattern: "exponential_backoff|retry"
    files: src/http_client.py
commit: "abc1234"  # 固定到特定 commit 以保证可复现
```

报告输出示意:

```
Task: add-retry-logic (3 runs each)
┌──────────────┬───────────┬────────┬────────┬─────────────┐
│ Agent        │ Pass Rate │ Cost   │ Time   │ Consistency │
├──────────────┼───────────┼────────┼────────┼─────────────┤
│ claude-code  │ 3/3       │ $0.12  │ 45s    │ 100%        │
│ aider        │ 2/3       │ $0.08  │ 38s    │  67%        │
└──────────────┴───────────┴────────┴────────┴─────────────┘
```

judge 各类型片段:

```yaml
# 确定性
judge:
  - type: pytest
    command: pytest tests/ -v
  - type: command
    command: npm run build
# 模式匹配
  - type: grep
    pattern: "class.*Retry"
    files: src/**/*.py
# LLM 评审
  - type: llm
    prompt: |
      Does this implementation correctly handle exponential backoff?
      Check for: max retries, increasing delays, jitter.
```

## 注意事项

- **安装前先审源**:agent-eval 从仓库安装,安装前确认源码可信再用。
- **git worktree 隔离**:每次运行独立 worktree,免 Docker;但要确保基础 repo 干净、commit 已固定,否则不同 Agent 起点不一致会污染对比。
- **样本量**:runs=3 是底线;通过率接近时需加大 runs 才能区分(一致性低说明 Agent 不稳定,比平均通过率更值得警惕)。
- **成本不可得**:部分 Agent/接口无法回传费用,Cost 列可能为空,别据此误判。
- **judge 决定结论**:弱 judge(只 grep)会高估通过率;尽量用测试/构建这类确定性判据,LLM judge 仅兜底且需缓解其偏差(见 `llm-judge-evaluation`)。
- 评测结果是选型参考,不替代你对中选 Agent 产出的人工复核与环境验证。

## 互见

- related:`llm-judge-evaluation` —— 当 judge 用 `type: llm` 时,靠它做无偏、可校准的 LLM 评审。
- related:`llm-agent-benchmarking` —— 单个智能体的深度基准、对抗测试与回归门禁(本技能聚焦"多 Agent 横向对比",它聚焦"单 Agent 纵向可靠性")。
- related:`autonomous-coding-agent-patterns` —— 评测中发现短板后,用其改进 Agent 行为。
- combines_with:`parallel-agent-hub` —— 并行编排多个 Agent 运行,加速对决。
- combines_with:`llm-model-router` —— 用评测数据驱动"按任务选 Agent/模型"的路由策略。

---
采编自 affaan-m/everything-claude-code(MIT 许可);CLI 源自 github.com/joaquinhuigomez/agent-eval。

Attribution

findscripterfindscripter
View sourceSee grades on GitHubMore from findscripter →
SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments (0)

No comments yet. Be the first to comment!

SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Related Skills

Caveman

Terse caveman voice: answer first, fluff gone, every technical fact kept. Use for /caveman, "caveman mode", "talk like caveman", "be brief", "less tokens". Stays on until "stop caveman" or "normal mode".

1100021 votes

Hyperplan

Adversarial multi-agent planning skill. Self-orchestrates 5 hostile category members (unspecified-low, unspecified-high, deep, ultrabrain, artistry) via team-mode for ruthless cross-critique debate, distills only the defensible insights, then MANDATORILY hands the distilled insight bundle to the `plan` agent for executable plan formalization. Use when planning needs maximum rigor and surfacing of weak assumptions, blind spots, and over-engineering. Triggers: 'hyperplan', 'hpp', '/hyperplan', ...

698461 votes

Writing Skills

Create and manage Claude Code skills in HASH repository following Anthropic best practices. Use when creating new skills, modifying skill-rules.json, understanding trigger patterns, working with hooks, debugging skill activation, or implementing progressive disclosure. Covers skill structure, YAML frontmatter, trigger types (keywords, intent patterns), UserPromptSubmit hook, and the 500-line rule. Includes validation and debugging with SKILL_DEBUG. Examples include rust-error-stack, cargo-dep...

3931 votes

Mcp Code Execution

Routes multi-tool workflows through MCP servers for large datasets and pipelines. Use when Bash tool overhead is limiting throughput on data-heavy tasks.

3421 votes

catchup

Recovers the conversation and failed tool calls of a previous Codex, Amp, Claude Code, Antigravity, Cline, Copilot CLI, Cursor, DeepSeek Harness, Grok Build, Kimi, OpenCode, Pi Agent, or ZCode session. Use when the user says "catch up", "what did the last session do", "get me up to speed", "I switched agents", asks to recover/summarize a previous session before continuing, or asks to diagnose or report a catchup failure. Do NOT use for the current conversation, git history, or any non-agent log.

741 votes
View all in ai-agents →