The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/fetch_xhs_trends.py utilizes subprocess.run to execute PowerShell commands on Windows systems to retrieve environment variables. Additionally, the skill instructions in SKILL.md direct the agent to execute shell commands (export on macOS/Linux and [Environment]::SetEnvironmentVariable on Windows) to modify the operating environment.
[COMMAND_EXECUTION]: The skill instructs the agent to automatically modify user shell profile files, specifically ~/.zshrc and ~/.bashrc, by appending export commands. This is a persistence mechanism that modifies system startup behavior.
[CREDENTIALS_UNSAFE]: The skill encourages storing the REDFOX_API_KEY in plain text within shell configuration files and environment variables. While intended for functional configuration, storing secrets in files like .bashrc or .zshrc makes them accessible to any process reading these common configuration files and may expose them in shell history.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to https://redfox.hk/story/api/cozeSkill/getXhsCozeSkillData using the requests library to fetch trending data. While this is a vendor-owned domain, it represents an external data dependency.
[INDIRECT_PROMPT_INJECTION]: The skill has a high surface for indirect prompt injection as it fetches external data from an API and interpolates it directly into the agent's context to generate and score titles. There are no explicit boundary markers or sanitization logic to prevent the fetched data from influencing the agent's behavior beyond the intended scope.
[DATA_EXFILTRATION]: User-provided keywords and the system's REDFOX_API_KEY are transmitted to the external domain redfox.hk during data retrieval operations.

xiaohongshu-title-score