The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches interview experience posts and question banks from external platforms including GitHub (github.com) and NowCoder (nowcoder.com). These network operations are necessary for the skill's declared purpose of gathering real-world interview data from public repositories and discussion forums.
[COMMAND_EXECUTION]: The skill uses subprocess.run in scripts/scrape/mediacrawler_driver.py to invoke the MediaCrawler command-line tool for Xiaohongshu searches. The execution is handled safely by passing a list of arguments and avoiding shell execution (shell=False), which mitigates command injection risks.
[DATA_EXPOSURE]: The skill accesses local resume files provided by the user (supporting PDF, text, and images via Vision/OCR). This data is processed locally to generate search queries and anchor interview questions to the user's projects. The analysis found no evidence of this sensitive information being transmitted to unauthorized third-party servers.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from the web (interview posts). To mitigate potential prompt injection risks, it uses a deterministic Python script (scripts/corpus/extraction.py) to segment the text into specific question "candidates" before the agent processes them. This structure helps prevent malicious instructions embedded in crawled content from overriding the agent's behavior.

interview-radar