The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content retrieved from external search engines, which creates a potential surface for indirect prompt injection attacks where malicious instructions hidden in web pages could influence agent behavior.
Ingestion points: The scripts/search.sh script uses curl to fetch live data from Baidu, Google, Bing, and DuckDuckGo based on user queries.
Boundary markers: There are no explicit boundary markers or instructions telling the agent to treat the fetched search results as untrusted content.
Capability inventory: The skill possesses network access capabilities via curl and local execution capabilities via python3.
Sanitization: The script includes basic sanitization that uses sed to strip HTML tags from the retrieved search results before they are echoed to the agent.
[DYNAMIC_EXECUTION]: The skill uses a dynamic execution pattern to perform URL encoding by interpolating user-provided strings into a Python command.
Evidence: The encode_url function in scripts/search.sh calls python3 -c "import urllib.parse; print(urllib.parse.quote('$input', safe=''))".
Mitigation: This execution pattern is significantly mitigated by the sanitize_input function, which is called prior to encoding and removes single quotes and other special characters that could be used to break out of the Python string literal.

curl-search