The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from Dianping.com using tools like readability and snapshot. This introduces a surface for indirect prompt injection where malicious instructions embedded in merchant descriptions, user reviews, or tags could influence the agent's behavior.
Ingestion Points: SKILL.md (via browser tool actions), references/deep-info-extraction.md.
Boundary Markers: The skill utilizes the readability tool, which performs basic content cleaning, but no specific boundary markers or "ignore" instructions are defined for the extracted data.
Capability Inventory: The skill instructions involve browser navigation, element interaction, and structured data extraction. The package also includes Python scripts capable of file system operations.
Sanitization: Standard browser tool cleaning is present; however, no advanced sanitization of the scraped text is documented.
[COMMAND_EXECUTION]: The skill package includes local Python scripts (scripts/package_skill.py and scripts/quick_validate.py). While these are standard developer tools for skill management, they represent executable code that the agent or user may run in the local environment.
Evidence: scripts/package_skill.py performs file system writes (creating ZIP files) and scripts/quick_validate.py parses YAML configuration.
[SAFE]: The skill demonstrates high-quality instructional design, including clear guidance for the agent to pause and wait for user interaction during login events, which enhances transparency and safety.

dianping-info-query