Add feapder crawler Codex skill

[ShellMonster]

Summary

Add a Codex Skill for feapder crawler development under Skills/feapder-crawler.

The skill packages feapder usage guidance for AI agents, covering:

• spider selection: AirSpider, Spider, TaskSpider, BatchSpider
• standard project workflow: feapder create -p <project_name> and feapder create -s <spider_name>
• existing-project handling: locate the real project root and merge changes into the active setting.py
• request/response patterns, item pipelines, parser integration, rendering, proxy, dedup, retry, and failure hooks
• feaplat deployment diagnosis and commonly used feapder utilities
• guardrails that require agents to stay on feapder paths unless the user explicitly authorizes a fallback

The guidance text is written in Chinese, while code, class names, commands, and config identifiers remain in English.

Testing

• Ran Codex Skill validation:

  python /Users/daozhang/.codex/skills/.system/skill-creator/scripts/quick_validate.py /Users/daozhang/Downloads/feapder/Skills/feapder-crawler

  Result: Skill is valid!

• Ran approximately 76 Codex sub-agent evaluation cases while developing and tuning the skill:

  • 45 trigger and boundary cases:

    • 15 explicit feapder should-trigger prompts
    • 15 contextual should-trigger prompts where the project/dependency context uses feapder
    • 15 should-not-trigger prompts for requests-only, Scrapy, Playwright-only, FastAPI, Celery, BeautifulSoup, SQLAlchemy, aiohttp, and other non-feapder tasks
    • Result: 45/45 matched expected trigger behavior

  • 10 output adherence prompts, checked across 20 expected behaviors:

    • AirSpider crawler generation
    • CsvPipeline and ITEM_PIPELINES
    • BatchSpider task source handling
    • Redis/settings troubleshooting
    • JavaScript rendering through feapder-supported paths
    • curl/debug conversion
    • Scrapy boundary handling
    • feapder project requesting direct requests usage
    • failed request handling
    • plain Python crawler boundary behavior
    • Result: 20/20 followed expected feapder guidance

  • 8 authorization and fallback cases:

    • Verified the agent must explain reason, impact, and alternatives, then ask the user for authorization before falling back to requests, Scrapy, standalone Playwright, handwritten CSV, or direct SQL
    • Result: 8/8 passed

  • 8 project creation and settings cases:

    • Existing project setting.py incremental edits
    • Preserving existing ITEM_PIPELINES
    • New project creation with feapder create -p
    • Existing standard project spider creation from <project_root>/spiders/ using feapder create -s
    • Single-file AirSpider boundary handling
    • Result: 16/16 expected checks passed

  • Focused retests after tuning:

    • feaplat read-only diagnosis
    • parser integration templates
    • TaskSpider MySQL/Redis task source templates
    • failure hooks such as exception_request and failed_request
    • project root and CLI execution directory rules
    • Result: 10/10 focused checks passed

[ShellMonster]

补充一个这个 Skill 能解决的具体例子：

feapder 的 Response.json 和 requests.Response.json() 不一样。源码里 json 是 @property，文档也写了 feapder 应该使用：

data = response.json

而不是：

data = response.json()

但如果没有这个 Skill，AI 即使读过源码和文档，也很容易被 requests 的通用习惯带偏，在生成“请求某个 JSON API 并解析字段”的爬虫代码时写成 response.json()，导致运行时报错。

这个 PR 里的 Skill 把这类 feapder 专属 API 差异收进了 references/request-response.md，并在测试里专门验证了 JSON API 解析场景：明确 feapder、现有 feapder 项目、以及用户没明说 feapder 但项目上下文是 feapder 的 3 类 case，生成代码都使用了 response.json，没有退化成 requests 风格。