The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external YouTube search results which could potentially contain malicious instructions aimed at influencing the agent's downstream behavior.
Ingestion points: The response from https://ai-factory.frevana.com/service/youtube-search is processed in scripts/search_youtube.sh.
Boundary markers: None are defined in the instructions to delimit external data from agent instructions.
Capability inventory: Network access (curl), shell command execution (bash, python3), and file system write access (cp, mkdir).
Sanitization: The script validates that the API response is valid JSON but does not sanitize or filter the textual content within the JSON fields.
[DATA_EXFILTRATION]: The skill transmits user-provided search queries and an authentication token (FREVANA_TOKEN) to an external service managed by the vendor.
Evidence: scripts/search_youtube.sh uses curl to send a POST request containing the search query and the Bearer token to https://ai-factory.frevana.com/service/youtube-search.
[COMMAND_EXECUTION]: The skill utilizes a bash script to coordinate network requests and can be directed to write output to local files.
Evidence: scripts/search_youtube.sh executes shell commands including curl and python3. It also uses mkdir -p and cp to save search results to a user-specified path provided via the --output argument.

youtube-search