The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill includes instructions that enable the deserialization of potentially untrusted data. * Evidence: In SKILL.md, the LangChain integration section demonstrates loading a local index using allow_dangerous_deserialization=True. This setting permits the use of Python's pickle module, which is inherently unsafe and can lead to arbitrary code execution if the index file originates from an untrusted source.
[INDIRECT_PROMPT_INJECTION]: The skill processes external content, establishing a surface for indirect prompt injection. * Ingestion points: The skill ingests external docs and user-provided query strings in the LangChain and LlamaIndex examples within SKILL.md. * Boundary markers: There are no specific instructions for using delimiters or boundary markers to separate untrusted data from the agent's primary instructions. * Capability inventory: The skill performs file I/O (saving and loading indices) and vector search operations across SKILL.md and references/index_types.md. * Sanitization: The provided code snippets do not include logic for sanitizing or validating external input before it is processed by the search engine.

faiss