The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from external, untrusted sources by providing instructions to scrape report URLs. This introduces a surface area for indirect prompt injection, where malicious content embedded in the source reports could potentially attempt to override agent instructions.
Ingestion points: The skill uses context_dev_scrape_markdown and context_dev_extract_structured_data to process content from user-supplied report URLs, as defined in references/sandbase-api-map.md.
Boundary markers: The instructions do not define specific delimiters or safety prompts to isolate the scraped content from the agent's core analysis logic.
Capability inventory: The skill utilizes sandbase_call_tool to interact with financial data and provides extensive logic for narrative generation and management reporting.
Sanitization: There are no documented procedures for sanitizing or validating the contents of the report URLs before the data is integrated into the prompt context.

variance-analysis