The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The scripts scripts/check_sensitive_words.py and scripts/fetch_xhs_trends.py are designed to automatically harvest the REDFOX_API_KEY by scanning sensitive shell configuration files in the user's home directory, including .zshrc, .bashrc, .bash_profile, .profile, and .zprofile. Accessing these files is generally considered high-risk as it could potentially expose other environment variables or system configurations stored in those files.
[DATA_EXFILTRATION]: User-supplied content, including drafted notes, text extracted from uploaded documents (DOCX, TXT), and content scraped from remote URLs, is transmitted to external endpoints at https://redfox.hk/story/api/... for compliance checking and analysis. While this is consistent with the skill's stated purpose, it involves sending user data to a third-party server.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from local files and remote websites, creating a potential surface for indirect prompt injection attacks.
Ingestion points: scripts/check_sensitive_words.py reads content from user-uploaded files and processes arbitrary external URLs.
Boundary markers: There is no evidence of strict delimiters or instructions to the model to ignore potential instructions embedded in the ingested content.
Capability inventory: The skill possesses network capabilities (requests, urllib) and file reading capabilities (python-docx).
Sanitization: Standard HTML parsing is performed, but no specific sanitization for malicious AI instructions is implemented.
[COMMAND_EXECUTION]: The skill utilizes subprocess.run to execute shell commands, specifically for managing browser dependencies (playwright install chromium) and interacting with Windows environment variables via PowerShell. While these are used for setup and configuration, they demonstrate a capability for system-level command execution.

xiaohongshu-creator