The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's workflow in SKILL.md involves executing a Python script via the shell: python3 scripts/search_xhs.py "<关键词>". If the agent performs direct string interpolation of user-supplied keywords into this shell command without proper sanitization or escaping, it presents a risk of command injection.
[PROMPT_INJECTION]: The skill ingests and analyzes content from external social media notes (titles and descriptions) fetched via an API. This creates an attack surface for indirect prompt injection, where carefully crafted text in a social media post could attempt to override the agent's instructions during the analysis phase.
Ingestion points: Untrusted data items from the search API are processed and summarized within SKILL.md.
Boundary markers: None present; the agent is instructed to process the raw JSON items directly.
Capability inventory: The skill includes network access capabilities via the urllib library in scripts/search_xhs.py.
Sanitization: No sanitization or filtering of the external note content is specified before the agent processes it for "选题信号" (topic signals).
[DATA_EXFILTRATION]: The script scripts/search_xhs.py performs network POST requests to doubaoya.com. While this is the core functionality of the skill, it involves sending user-derived search terms and an authentication token to an external domain.

xiaohongshu-search