The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it retrieves and processes external content from the internet without sufficient safeguards.
Ingestion points: The scripts/tavily.sh script fetches search results and raw web content via the Tavily API in the search and extract functions.
Boundary markers: Absent. The skill does not use delimiters or instructions to help the agent distinguish between its own logic and the untrusted data retrieved from the web.
Capability inventory: The skill has broad capabilities including Bash, Read, Write, and Edit as listed in the allowed-tools section of SKILL.md.
Sanitization: Absent. The script processes the API response using jq but does not filter or sanitize the text content for potential malicious instructions before providing it to the agent.
[COMMAND_EXECUTION]: The script scripts/tavily.sh uses unsafe string interpolation when constructing the JSON payload for API requests.
Evidence: In both the search and extract functions, shell variables $query and $max_results are directly interpolated into the JSON string passed to the curl command (e.g., "query": "$query"). If a user-provided query contains escaped double quotes or other JSON control characters, it could be used to manipulate the API request structure.

tavily