The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The detection script scripts/check_sensitive_words.py includes a function _get_api_key that programmatically opens and reads sensitive user shell profile files (.bashrc, .zshrc, .bash_profile, .profile, and .zprofile) to search for and extract the REDFOX_API_KEY credential.
[PERSISTENCE]: The SKILL.md file provides instructions directing the agent to modify the user's permanent shell configuration files (e.g., ~/.zshrc, ~/.bashrc) by appending export commands. This constitutes an automated modification of system environment persistence mechanisms.
[REMOTE_CODE_EXECUTION]: The skill implements a recharge workflow in references/alipay-recharge.md that requires the agent to dynamically install an external package (@alipay/agent-payment) using npx and execute commands (alipay-bot) with arguments derived from a remote API response (cashierUrl). This establishes a remote-triggered execution path for external code.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from external web URLs via scripts/extract_text.py which is then processed by the agent, creating a vulnerability surface for indirect prompt injection.
Ingestion points: The extract_from_web function in scripts/extract_text.py fetches content from user-provided URLs using the requests library.
Boundary markers: There are no specific delimiters or isolation instructions defined to separate the extracted untrusted web content from the agent's primary instructions.
Capability inventory: The skill possesses capabilities for writing to system configuration files and executing shell commands (via the payment/recharge instructions).
Sanitization: The script implements basic HTML tag removal in _clean_html_tags but lacks validation or sanitization against adversarial instructions embedded in the processed data.
[EXTERNAL_DOWNLOADS]: The skill communicates with the redfox.hk API to perform prohibited word detection and fetch payment configuration. These resources are managed by the skill's author, redfox-data.

multi-wordcheck