The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a core Python script (scripts/check_sensitive_words.py) to handle file extraction, web scraping, and API communication.
[EXTERNAL_DOWNLOADS]: The skill requires standard Python libraries (requests, python-docx, beautifulsoup4, playwright) and installs the Chromium browser for web content extraction.
[DATA_EXFILTRATION]: Extracted user content is sent to the vendor's API at https://redfox.hk/story/api/cozeSkill/sensitiveWordSearch for processing. This is the skill's primary function and targets the author's own infrastructure.
[CREDENTIALS_UNSAFE]: The Python script automatically reads the user's shell configuration files (e.g., .bashrc, .zshrc) to retrieve the REDFOX_API_KEY. This is intended for seamless configuration but involves accessing sensitive system files.
[PROMPT_INJECTION]: The skill processes untrusted data from external websites and uploaded files, creating a surface for indirect prompt injection attacks.
Ingestion points: extract_from_file and extract_from_web functions in scripts/check_sensitive_words.py.
Boundary markers: No explicit delimiters or isolation instructions are provided to the agent when processing the fetched text.
Capability inventory: The agent can execute scripts, perform network operations to the vendor's API, and write optimized text to local files.
Sanitization: Content is parsed for text extraction, but the skill lacks mechanisms to prevent the agent from executing instructions potentially hidden within the scraped content.

wechat-prohibited-word