The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes numerous examples of prompt injection payloads, such as "Ignore previous instructions" and "You are now DAN," but these are explicitly categorized as "Testing Patterns" intended for security researchers to use during application audits. They are not used as instructions for the agent itself.
[COMMAND_EXECUTION]: The skill defines allowed-tools including Bash, Grep, and Glob. These are used within the context of security inventory and architecture reviews. No malicious or automated destructive command patterns were found in the instructional body.
[DATA_EXFILTRATION]: No network-capable tools (like WebFetch) are allowed, and no patterns indicating the transmission of sensitive data to external servers were detected.
[SAFE]: The skill does not include any external dependencies, remote scripts, or obfuscated content. It functions as a structured knowledge base for security best practices.

security-app-llm