Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

/search-first — 编码前先研究 · affaan-m | DecimalAI

affaan-m/search-first

SkillScore91Maintained

Research-before-coding workflow. Search for existing tools, libraries, and patterns before writing custom code. Systematizes the "search for existing solutions before implementing" approach. Use when starting new features or adding functionality.

researchby affaan-m

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Coding pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/affaan-m-search-first/SKILL.md

decimalai skills pull affaan-m-search-first --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/affaan-m-search-first/SKILL.md https://app.decimal.ai/s/affaan-m-search-first/SKILL.md

Raw file: /s/affaan-m-search-first/SKILL.md ↗ · machine-readable summary at /s/affaan-m-search-first.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 47d ago

Yes — +14% lift · at higher cost+56% tokens0% turns91% of whole cases pass with skill (20/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-18	✗→✓	▲ Improved	29%	0%
case-19	✗→✓	▲ Improved	13%	0%
case-22	✗→✓	▲ Improved	213%	0%
case-01	✓→✓	＝ Same ✓	46%	0%
case-02	✓→✓	＝ Same ✓	49%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

affaan-m / ecc

.kiro/skills/search-first/SKILL.md

View on GitHub ↗

License MITImported 8/3/2026Version v6

Skill contents

SKILL.md

/search-first — 编码前先研究

系统化“在实现之前先寻找现有解决方案”的工作流程。

触发时机

在以下情况使用此技能：

开始一项很可能已有解决方案的新功能
添加依赖项或集成
用户要求“添加 X 功能”而你准备开始编写代码
在创建新的实用程序、助手或抽象之前

工作流程

┌─────────────────────────────────────────────┐
│  1. 需求分析                               │
│     确定所需功能                          │
│     识别语言/框架限制                     │
├─────────────────────────────────────────────┤
│  2. 并行搜索（研究员代理）                │
│     ┌──────────┐ ┌──────────┐ ┌──────────┐  │
│     │  npm /   │ │  MCP /   │ │  GitHub / │  │
│     │  PyPI    │ │  技能    │ │  网络     │  │
│     └──────────┘ └──────────┘ └──────────┘  │
├─────────────────────────────────────────────┤
│  3. 评估                                   │
│     对候选方案进行评分（功能、维护、      │
│     社区、文档、许可证、依赖）            │
├─────────────────────────────────────────────┤
│  4. 决策                                   │
│     ┌─────────┐  ┌──────────┐  ┌─────────┐  │
│     │  采用   │  │  扩展    │  │  构建   │  │
│     │ 原样    │  │  /包装   │  │  定制   │  │
│     └─────────┘  └──────────┘  └─────────┘  │
├─────────────────────────────────────────────┤
│  5. 实施                                   │
│     安装包 / 配置 MCP /                    │
│     编写最小化自定义代码                   │
└─────────────────────────────────────────────┘

决策矩阵

| 信号 | 行动 | |--------|--------| | 完全匹配，维护良好，MIT/Apache 许可证 | 采纳 — 直接安装并使用 | | 部分匹配，基础良好 | 扩展 — 安装 + 编写薄封装层 | | 多个弱匹配 | 组合 — 组合 2-3 个小包 | | 未找到合适的 | 构建 — 编写自定义代码，但需基于研究 |

使用方法

快速模式（内联）

在编写实用程序或添加功能之前，在脑中过一遍：

这已经在仓库中存在吗？ → 首先通过相关模块/测试检查 rg
这是一个常见问题吗？ → 搜索 npm/PyPI
有对应的 MCP 吗？ → 检查 ~/.claude/settings.json 并进行搜索
有对应的技能吗？ → 检查 ~/.claude/skills/
有 GitHub 上的实现/模板吗？ → 在编写全新代码之前，先运行 GitHub 代码搜索以查找维护中的开源项目

完整模式（代理）

对于非平凡的功能，启动研究员代理：

任务（子代理类型="通用型"，提示="
  研究现有工具用于：[描述]
  语言/框架：[语言]
  约束：[任何]

  搜索：npm/PyPI、MCP 服务器、Claude Code 技能、GitHub
  返回：结构化对比与推荐
")

按类别搜索快捷方式

开发工具

Linting → eslint, ruff, textlint, markdownlint
Formatting → prettier, black, gofmt
Testing → jest, pytest, go test
Pre-commit → husky, lint-staged, pre-commit

AI/LLM 集成

Claude SDK → 使用 Context7 获取最新文档
提示词管理 → 检查 MCP 服务器
文档处理 → unstructured, pdfplumber, mammoth

数据与 API

HTTP 客户端 → httpx (Python), ky/got (Node)
验证 → zod (TS), pydantic (Python)
数据库 → 首先检查是否有 MCP 服务器

内容与发布

Markdown 处理 → remark, unified, markdown-it
图片优化 → sharp, imagemin

集成点

与规划器代理

规划器应在阶段 1（架构评审）之前调用研究员：

研究员识别可用的工具
规划器将它们纳入实施计划
避免在计划中“重新发明轮子”

与架构师代理

架构师应向研究员咨询：

技术栈决策
集成模式发现
现有参考架构

与迭代检索技能

结合进行渐进式发现：

循环 1：广泛搜索 (npm, PyPI, MCP)
循环 2：详细评估顶级候选方案
循环 3：测试与项目约束的兼容性

示例

示例 1：“添加死链检查”

需求：检查 Markdown 文件中的失效链接
搜索：npm "markdown dead link checker"
发现：textlint-rule-no-dead-link（评分：9/10）
行动：采纳 — npm install textlint-rule-no-dead-link
结果：无需自定义代码，经过实战检验的解决方案

示例 2：“添加 HTTP 客户端包装器”

需求：具备重试和超时处理能力的弹性 HTTP 客户端
搜索：npm "http client retry"、PyPI "httpx retry"
发现：got（Node）带重试插件、httpx（Python）带内置重试功能
行动：采用——直接使用 got/httpx 并配置重试
结果：零定制代码，生产验证的库

示例 3：“添加配置文件 linter”

需求：根据模式验证项目配置文件
搜索：npm "config linter schema"、"json schema validator cli"
发现：ajv-cli（评分：8/10）
操作：采用 + 扩展 —— 安装 ajv-cli，编写项目特定的模式
结果：1 个包 + 1 个模式文件，无需自定义验证逻辑

反模式

直接跳转到编码：不检查是否存在就编写实用程序
忽略 MCP：不检查 MCP 服务器是否已提供该能力
过度定制：对库进行如此厚重的包装以至于失去了其优势
依赖项膨胀：为了一个小功能安装一个庞大的包

Illustrative Want to see what affaan-m/search-first would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/11/2026, 19:41 UTC · 22 cases

Lift

+14%

91% vs 77%

Agent turns

1.0 vs 1.0

Tokens

+56%

2,692.82 vs 1,726.09

fail → pass · 3pass → pass · 17fail → fail · 2

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 We need to deeply flatten a nested JavaScript array `[1, [2, [3, [4]]]]` in Node.js 18. Should we search external package registries for an npm utility package like `lodash.flatten` or `array-flatten`, or how should we solve this task?	pass→pass	8,009	4,959	-38%	1	1	0%	1,589	2,316	+46%	0	0	—
case-02 We need to implement our company's internal custom pricing tiers based on confidential client contract terms and proprietary margin rules. Should we search external package registries like npm or PyPI for an existing open-source library to handle this pricing calculation?	pass→pass	10,529	7,242	-31%	1	1	0%	1,837	2,744	+49%	0	0	—
case-03 In a Node.js 20 backend service, we need to combine path segments `src` and `controllers` into a platform-agnostic file path. Should we search external package registries for a third-party path utility package, or what built-in approach should we use?	pass→pass	6,491	5,012	-23%	1	1	0%	1,252	2,494	+99%	0	0	—
case-04 We need a CLI tool to scan Markdown files in our repository and verify that all hyperlinks are active and returning HTTP 200. We are tempted to write a custom Python script using regex and requests. What package should we adopt instead?	fail→fail	11,629	12,556	+8%	1	1	0%	1,966	3,626	+84%	0	0	—
case-05 We are building a Python backend service calling external APIs. We need native async support along with built-in HTTP retry and timeout handling so we don't have to code retry loops around urllib. Which Python HTTP library should we select?	pass→pass	13,513	13,237	-2%	1	1	0%	2,648	4,009	+51%	0	0	—
case-06 In a Node.js project, we need an HTTP request client that natively supports retries and timeouts out of the box without writing custom wrapper logic around fetch(). Which Node library should we choose?	pass→pass	9,739	10,138	+4%	1	1	0%	1,877	3,450	+84%	0	0	—
case-07 We need runtime data validation and schema declaration for incoming API requests in TypeScript. Rather than writing custom validation functions, which library should we choose?	pass→pass	12,829	8,439	-34%	1	1	0%	2,294	3,012	+31%	0	0	—
case-08 We need structured data validation, parsing, and settings management in Python. Rather than building custom validation methods, which Python library should we select?	pass→pass	10,457	8,156	-22%	1	1	0%	2,001	2,892	+45%	0	0	—
case-09 We need to parse Markdown documents into an Abstract Syntax Tree (AST) and run custom transformations in Node.js. What library ecosystem should we use instead of writing custom regex parsers?	pass→pass	10,801	8,865	-18%	1	1	0%	2,114	3,088	+46%	0	0	—
case-10 We need to extract structured tables and text from PDF documents in Python. Should we write custom binary parsing logic from scratch, or which Python library should we adopt?	pass→pass	12,739	9,708	-24%	1	1	0%	2,322	3,203	+38%	0	0	—
case-11 We need high-performance image resizing and WebP conversion in a Node.js backend. Which library should we use instead of invoking external shell commands?	pass→pass	8,418	7,950	-6%	1	1	0%	1,715	2,880	+68%	0	0	—
case-12 We want to replace slow, separate Python linters and formatters with a single fast Rust-based tool. Which tool should we adopt?	pass→pass	8,133	7,355	-10%	1	1	0%	1,537	2,896	+88%	0	0	—
case-13 We want to automatically run formatters and linters on staged files before each Git commit in a Node repository. Instead of hand-writing custom shell scripts in .git/hooks/, which packages should we combine?	pass→pass	8,088	6,350	-21%	1	1	0%	1,534	2,676	+74%	0	0	—
case-14 We found an open-source library that meets 100% of our feature specifications, is actively maintained, and uses the MIT license. What decision action should we take?	pass→pass	7,126	2,622	-63%	1	1	0%	1,277	1,862	+46%	0	0	—
case-15 We found an open-source package that provides 80% of required functionality with a solid codebase, but lacks one internal business rule. What decision action should we take?	pass→pass	10,454	5,517	-47%	1	1	0%	1,755	2,498	+42%	0	0	—
case-16 No single package solves our entire requirement, but two small, focused, well-maintained packages each cleanly solve distinct parts of the problem. What decision action should we take?	pass→pass	7,703	4,672	-39%	1	1	0%	1,355	2,246	+66%	0	0	—
case-17 Before searching npm, PyPI, or GitHub for a utility helper function, what initial search step should be performed locally in the repository?	fail→fail	4,022	2,601	-35%	1	1	0%	701	1,885	+169%	0	0	—
case-18 We want to connect our development environment with an external database. Before building custom integration code, where should we check for existing local MCP capability?	fail→pass	10,915	5,310	-51%	1	1	0%	1,836	2,360	+29%	0	0	—
case-19 Before writing custom automation prompt workflows, where in the local filesystem should we check for existing assistant skills?	fail→pass	9,376	2,511	-73%	1	1	0%	1,620	1,823	+13%	0	0	—
case-20 For a major feature research task requiring parallel searching across registries, MCP servers, skills, and GitHub, what type of researcher sub-agent should be launched?	pass→pass	10,670	3,142	-71%	1	1	0%	1,863	2,060	+11%	0	0	—
case-21 We need to validate project configuration files against JSON Schema definitions in CI. We are considering writing a custom Node script. What CLI tool should we adopt instead?	pass→pass	11,884	9,014	-24%	1	1	0%	2,178	3,020	+39%	0	0	—
case-22 When starting pre-coding research for a new feature, a developer suggests querying external registries like npm and PyPI first. What search step should be performed before querying external package registries?	fail→pass	4,371	4,242	-3%	1	1	0%	703	2,202	+213%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +14 percentage points is the difference between those two pass rates over the 22 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 77.3%
Pass rate with the skill: 90.9%
Cases attempted: 22
Cases counted toward lift: 22
Cases that regressed: 0
Skill version measured: v6
Run completed: 8/11/2026, 19:41 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-18

Prompt

We want to connect our development environment with an external database. Before building custom integration code, where should we check for existing local MCP capability?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run91

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+52% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 47 days ago

Other measured skills in the registry, with their headline benchmark lift.