The Agent Skills Directory

[PROMPT_INJECTION]: The skill acts as an ingestion point for untrusted data from the web, which constitutes an indirect prompt injection surface.
Ingestion points: Untrusted data from web search results is fetched via the Skywork API in scripts/web_search.py and written to local text files.
Boundary markers: The script uses basic markers like [result-n] to separate content, but there are no explicit delimiters or instructions to the agent to ignore malicious commands embedded in the results.
Capability inventory: The script scripts/web_search.py performs network requests to api-tools.skywork.ai and writes results to files in a temporary directory (tempfile.mkdtemp); the agent then reads these files to process the data.
Sanitization: The script performs standard JSON parsing and whitespace stripping but does not sanitize the search result content for embedded natural language instructions.

Skywork Search