The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is designed to ingest and extract information from untrusted user-uploaded documents such as protocols, manuscripts, and data extraction spreadsheets. Malicious instructions embedded in these documents could potentially influence the agent's behavior during the drafting phase.
Ingestion points: Documents uploaded in .docx, .pdf, .xlsx, or spreadsheet formats (SKILL.md).
Boundary markers: The instructions do not define specific delimiters or "ignore" instructions when processing the content of these external files.
Capability inventory: The skill utilizes web_search for reference verification and executes a local Python validation script (SKILL.md).
Sanitization: No explicit sanitization or input validation logic is described for the content extracted from uploaded documents.
[COMMAND_EXECUTION]: The skill instructions specify the execution of a local script, python scripts/office/validate.py, to validate generated Word documents before presenting them to the user (SKILL.md). This is a standard internal validation procedure.

slr-prisma