The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text in the form of vulnerability descriptions, which constitutes a surface for indirect prompt injection.
Ingestion points: The description input in .calcifier/recipe.py is passed directly into the agent context.
Boundary markers: The skill relies on natural language instructions rather than strict delimiters to separate the input data from the core instructions.
Capability inventory: There are no exploitable capabilities; the script is restricted to reading local reference files and emitting a text-based report. It cannot write to the filesystem, execute system commands, or perform network operations.
Sanitization: The input is not sanitized or filtered before being processed by the underlying model.

cvss-scoring