The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and analyzes content from a wide range of external sources, including government portals, NGO reports, and non-traditional websites. This external data is untrusted and could contain adversarial prompts aimed at influencing the agent's behavior during data verification or processing.
Ingestion points: Web content retrieved via web_fetch across all search categories (A-G) and deep-dive searches.
Boundary markers: No specific delimiters or instructions to ignore embedded commands within the fetched data are defined in the skill instructions.
Capability inventory: The agent has capabilities to perform network requests, write files, and execute Python or Stata code for data processing.
Sanitization: The instructions do not specify any sanitization or validation steps for the retrieved external content before it is processed or used in script generation.
[DYNAMIC_EXECUTION]: The skill assists in generating and running Python code for web scraping (using libraries like Selenium and BeautifulSoup) and data inspection. The generation and execution of code based on parameters or content derived from external, untrusted sources creates a potential vulnerability surface.
[EXTERNAL_DOWNLOADS]: The skill performs numerous network operations to verify datasets and fetch documentation from arbitrary third-party domains. Although many targets are official repositories (e.g., Census, World Bank), the broad search strategy for creative sources inherently includes non-standard and less-verifiable sources.

find-data