The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions attempt to override system-level configurations by explicitly directing the agent or the platform environment not to use safety/invocation constraints such as disable-model-invocation or allow_implicit_invocation: false.
[COMMAND_EXECUTION]: The skill utilizes shell commands (curl and uuidgen) to perform network requests and generate unique identifiers for API idempotency.
[EXTERNAL_DOWNLOADS]: The skill interacts with an external service, deepapi.co, to perform web searches, scrape specific websites, and query social media data (Twitter) for research purposes.
[INDIRECT_PROMPT_INJECTION]: The skill has an inherent attack surface for indirect prompt injection because its primary function is to ingest and process untrusted data from arbitrary third-party websites and social media platforms.
Ingestion points: External content enters the agent's context through the responses of the deepapi.co API, specifically from the web search, website scraping, and Twitter search endpoints.
Boundary markers: There are no instructions provided to use delimiters or specific safety markers to differentiate between the agent's instructions and the untrusted content retrieved from the web.
Capability inventory: The skill possesses network capabilities via curl and the ability to process and summarize complex external data.
Sanitization: The instructions do not define any sanitization, validation, or filtering processes for the data retrieved from external sources before it is analyzed by the agent.

online-shopping