The Agent Skills Directory

[SAFE]: The skill provides conceptual guidance on AI safety architectures, including input validation, permission layering, and refusal strategies. No security threats were identified.
[PROMPT_INJECTION]: The content discusses prompt injection defense techniques and provides examples of how to mitigate them. It does not contain any malicious directives aimed at overriding the agent's behavior.
[DATA_EXFILTRATION]: There are no commands or instructions for accessing sensitive files or performing network operations. No hardcoded credentials or API keys are present.
[REMOTE_CODE_EXECUTION]: The skill contains no code snippets, package installations, or remote script download patterns.
[OBFUSCATION]: No obfuscated text, encoded URLs, or hidden characters were detected in the file.
[INDIRECT_PROMPT_INJECTION]: The skill serves as a static template for safety design. It does not ingest or process untrusted external data and lacks capabilities (e.g., file writes or network calls) that could be exploited via injection.

safety-guardrails