The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a safety notice that proactively warns the agent against following instructions embedded in processed data. While the text mentions specific injection phrases like 'ignore previous instructions', they are used strictly for defining defensive boundaries and do not represent an attempt to override the agent's system instructions.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze potentially untrusted external inputs like security logs and threat reports. It mitigates the risk of indirect injection by instructing the agent to treat these inputs as data for analysis only, to redact sensitive information, and to avoid executing any commands or scripts found within the provided context.
[EXTERNAL_DOWNLOADS]: The skill refers to established and well-known security resources from organizations like MITRE, SigmaHQ, and Palantir. These references are used neutrally to provide technical guidance and do not include instructions for the agent to download or execute untrusted remote code.
[OBFUSCATION]: A Base64-encoded string is used within a sample Sigma rule as a test payload. This payload is benign, decoding to a simple 'echo' command, and is used appropriately to illustrate how to validate detection logic.

detection-engineering