The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process BIM element data from external files, which represents a surface for indirect instructions to be introduced into the agent context.
Ingestion points: Untrusted data is loaded via pd.read_excel as defined in SKILL.md and through user-provided files (CSV, Excel, JSON) mentioned in instructions.md.
Boundary markers: The instructions do not specify the use of delimiters or provide explicit directives to the agent to ignore natural language instructions that might be contained within element property fields.
Capability inventory: The skill's functions are restricted to spatial calculations, geometric intersection logic, and data summary generation. It does not contain any code for network operations, shell command execution, persistence, or sensitive file access.
Sanitization: Basic input sanitization is performed during the element loading process in SKILL.md through explicit type casting (converting coordinates to float and identifiers to str), which validates the structure of the incoming data.

bim-clash-detection