The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implementation in SKILL.md (Step 1) constructs a shell command using string interpolation of the filePath variable derived from user input. There is no evidence of path validation or sanitization before it is passed to RvtExporter.exe, which could allow for command injection if an attacker provides a malicious file path.
[INDIRECT_PROMPT_INJECTION]: The skill demonstrates a vulnerability surface where untrusted data from BIM elements is ingested and processed by an LLM.
Ingestion points: BIM element data (Category, Name, Volume, etc.) is read from Excel/CAD files in Step 2 and 3 of the pipeline implementation.
Boundary markers: The prompt template in Step 3 lacks delimiters or specific instructions to treat the element data as untrusted content, making the system susceptible to instructions embedded within BIM metadata.
Capability inventory: The skill has access to shell execution (RvtExporter.exe), filesystem operations (via n8n nodes), and network access to a vector database.
Sanitization: The implementation does not include any sanitization, filtering, or validation logic for the element data before it is interpolated into the LLM prompt.

n8n-cost-estimation