The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface by processing external data and possessing significant capabilities.
Ingestion points: Untrusted data enters the agent context through WebSearch, WebFetch, and get_sec_filing (for earnings call transcripts, investor decks, and press releases), as detailed in references/workflow.md (Steps 1 and 2).
Capability inventory: The skill has the capability to write .docx files using the docx internal skill and can modify user-supplied Excel workbooks using the model-update internal skill (referenced in references/workflow.md, Step 13). It also executes Python code (matplotlib, pandas) to generate charts.
Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the ingested external text, which increases the risk of the agent following instructions hidden inside financial documents.
Sanitization: The skill lacks automated sanitization, filtering, or escaping for the text it fetches and quotes in its deliverables.

earnings-analysis