The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs a local file path using user-supplied input ($ARGUMENTS) without explicit sanitization instructions. Specifically, it saves the final report to reports/{TICKER}earnings_prep{UPCOMING_CQ}.html. A malicious user could provide a ticker string containing path traversal characters (e.g., ../) to attempt writing the report to an unauthorized directory.
[PROMPT_INJECTION]: The skill relies on external data sources, including web searches and financial filings, to populate the report. This creates an indirect prompt injection vulnerability where malicious instructions embedded in those external sources could manipulate the agent's output or reasoning process. 1. Ingestion points: External web search results and SEC filings accessed via search_documents and WebSearch tools. 2. Boundary markers: None identified in the prompt to delineate or isolate untrusted data. 3. Capability inventory: File system write access (save report instruction). 4. Sanitization: No instructions provided to escape or validate the content retrieved from external sources before including it in the report.

earnings-prep