The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script (check_sensitive_words.py) to perform text extraction and coordinate the detection process.
[EXTERNAL_DOWNLOADS]: The skill fetches content from external URLs provided by the user. It also utilizes the Playwright library to download and run browser binaries for handling dynamic web content.
[DATA_EXFILTRATION]: User-provided data is transmitted to the vendor's domain (redfox.hk) for analysis. This network operation is the primary function of the skill and is required for word detection.
[CREDENTIALS_UNSAFE]: The core script includes logic to automatically scan the user's shell configuration files (such as ~/.bashrc, ~/.zshrc, and ~/.profile) to retrieve the REDFOX_API_KEY if it is not present in the environment variables.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from user-uploaded files and external web URLs. Evidence:
Ingestion points: Content is extracted via extract_from_file and extract_from_web in check_sensitive_words.py.
Boundary markers: None identified; there are no instructions to the agent to ignore instructions embedded within the processed text.
Capability inventory: Executes local scripts, performs network POST requests, and writes data to local .txt files.
Sanitization: Basic removal of script and style tags from HTML content using BeautifulSoup.

douyin-prohibited-word