The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill ingests untrusted data from external Google search results. While this presents an indirect prompt injection surface, the risk is minimized by the skill's design.
Ingestion points: Data enters the agent context via the JSON response from api.brightdata.com in scripts/search.sh.
Boundary markers: No explicit natural language delimiters are added to the output, but the data is returned as structured JSON.
Capability inventory: The skill is limited to curl (API request) and jq (JSON parsing). It has no file-write or arbitrary code execution capabilities.
Sanitization: The script uses jq to extract only specific fields (link, title, description), which prevents the raw, unparsed response from being passed to the agent.
[Data Exposure & Exfiltration] (SAFE): API credentials are correctly handled through environment variables (BRIGHTDATA_API_KEY). The script only sends this key to the official api.brightdata.com endpoint over HTTPS.
[Command Execution] (SAFE): User-provided search queries are safely URL-encoded using jq -sRr @uri before being interpolated into the search URL, preventing shell injection or command manipulation.

search