The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents numerous shell commands for environment setup and project initialization, such as pip install, playwright install, and scrapy startproject in SKILL.md. While these are instructional, the agent is directed to perform these actions which involve external command execution.
[DATA_EXFILTRATION]: The mitmproxy examples in SKILL.md (lines 515-533, 625-644) print intercepted request headers, request bodies, and save raw response content to disk. This constitutes a high risk of sensitive data exposure if the intercepted traffic contains authentication tokens, cookies, or personal user data.
[CREDENTIALS_UNSAFE]: The skill provides templates for API data collection (fetch_api_data in SKILL.md) that involve interpolating an api_key into the Authorization header. While no real keys are provided, the logic facilitates the handling of secrets in potentially logged or unencrypted contexts. It also includes a hardcoded connection string for a local Redis instance (redis://localhost:6379).
[EXTERNAL_DOWNLOADS]: The ProxyPool class in SKILL.md (lines 351-375) fetches a JSON file from raw.githubusercontent.com/proxifly/free-proxy-list/main/proxies/proxies.json. This downloads a mutable list of external proxy servers from an untrusted repository and immediately uses them to route scraper traffic, which could lead to man-in-the-middle attacks or traffic monitoring by malicious proxy providers.
[REMOTE_CODE_EXECUTION]: The skill recommends installing and using frida-tools and mitmproxy for App reverse engineering. These tools are capable of dynamic code injection and traffic manipulation, which, while standard for the stated task, represent high-privilege capabilities.
[INDIRECT_PROMPT_INJECTION]: The skill lacks explicit sanitization or boundary markers for the data it ingests from external websites (via requests, playwright, and scrapy). This presents a surface for indirect prompt injection where malicious instructions embedded in a scraped webpage could influence the agent's subsequent logic.
[ANTI_BOT_CIRCUMVENTION]: A significant portion of the skill is dedicated to bypassing security controls. It provides instructions for ddddocr for CAPTCHA solving, curl_cffi for WAF/Cloudflare avoidance, browser fingerprint concealment scripts, and explicitly advises setting ROBOTSTXT_OBEY = False in Scrapy settings.

web-scraper