The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses files in a sibling directory using relative paths (e.g., '../eva/'). While these appear to be vendor-owned resources, the behavior allows access to data outside the immediate skill package.
[PROMPT_INJECTION]: The skill processes untrusted user-supplied documents (marketing briefs) to generate structured instructions for other agent components. • Ingestion points: User input containing business briefs and requirements. • Boundary markers: None; the skill does not use delimiters to isolate untrusted data. • Capability inventory: Reading local file system and generating data for downstream tasks. • Sanitization: None; input is processed without validation for embedded instructions.

eva-brief