The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a reference to "ignore previous instructions" within its documentation of edge cases for Retrieval-Augmented Generation (RAG) systems. This is a descriptive example of an attack vector that the user might encounter in their own software, rather than an attempt to manipulate the AI agent's instructions.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill provides defensive guidance, instructing users to redact sensitive data from logs, avoid storing prompts in local storage, and ensure training logs do not contain customer data.
[REMOTE_CODE_EXECUTION]: The skill recommends standard security auditing tools such as cargo audit and cargo deny and emphasizes verifying model integrity using checksums (sha256sum). It also warns against tokenizers that download files from the internet during production as a runtime risk.

rust-ml-llm-debugging