The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection due to its processing of untrusted data from binaries. * Ingestion points: The skill reads metadata and content from the strings, funcs, and imports tables in the SKILL.md instructions. * Boundary markers: No special delimiters or instructions are used to separate binary-derived data from agent instructions. * Capability inventory: The skill execution environment has access to the Bash tool and the ability to modify the internal database using SQL UPDATE statements. * Sanitization: There is no evidence of validation or sanitization of strings or other identifiers retrieved from the analyzed binary.

xrefs