The Agent Skills Directory

[COMMAND_EXECUTION]: The execution instructions in SKILL.md (Step 2) interpolate the user-provided keyword directly into a shell command line: python search.py "关键词". If a user provides a keyword containing shell metacharacters such as semicolons, backticks, or quotes (e.g., "; rm -rf /; #), it could lead to arbitrary command execution on the host machine.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. 1. Ingestion points: Video titles, author names, and descriptions are scraped from the external Douyin website via scripts/search.py. 2. Boundary markers: Absent; the data is presented to the agent for analysis without escaping or delimiters. 3. Capability inventory: The agent has shell access and file-writing capabilities as documented in SKILL.md. 4. Sanitization: Absent; the scraped strings are used directly for "选题分析" (topic analysis), which could allow a malicious video title to influence agent behavior.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the DrissionPage library from PyPI during the setup phase described in SKILL.md.
[DATA_EXFILTRATION]: The script in scripts/search.py interacts with the local Google Chrome browser and may have access to the user's active login sessions, cookies, and profile data, which are then processed and stored in a temporary directory.

laohan-douyinsousuo