The Agent Skills Directory

[Command Execution] (SAFE): The script executes the curl binary via subprocess.run using a list of arguments (shell=False). This correctly prevents shell injection vulnerabilities, ensuring that search queries are treated strictly as data.
[Data Exposure & Exfiltration] (LOW): The skill performs network operations to api.duckduckgo.com and api.tavily.com. While these are necessary for search functionality, these domains are not on the trusted whitelist for exfiltration analysis. No sensitive local files are accessed or transmitted.
[Indirect Prompt Injection] (LOW): This skill ingests untrusted data from the internet. Ingestion points: API responses in scripts/web_search.py. Boundary markers: None. Capability inventory: Only performs network searches and returns results; has no file-write or system-modifying capabilities. Sanitization: Results are parsed into structured JSON.
[Credentials Unsafe] (SAFE): The skill avoids hardcoded secrets by retrieving the Tavily API token from the TRAVILY_TOKEN environment variable.

web-search