The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process project data from external files such as CSV, Excel, and JSON, as outlined in instructions.md. This capability introduces a surface for indirect prompt injection if those files contain malicious instructions disguised as data.
Ingestion points: Data provided by the user in Excel, CSV, or JSON formats as specified in the Input Format section of instructions.md.
Boundary markers: The instructions lack specific boundary markers or explicit guidelines for the agent to ignore natural language instructions embedded within the data values.
Capability inventory: The skill includes file system write permissions (filesystem in claw.json) and the ability to generate and save Excel files through the export_to_excel method in SKILL.md.
Sanitization: While the instructions mandate validating inputs, there are no specific technical sanitization methods provided to filter or escape instructions embedded in the external data sources.

carbon-calculator