The Agent Skills Directory

[SAFE]: The skill is strictly defensive in nature, providing a methodology for engineers to identify risks like model poisoning and insecure serialization in their own systems.
[SAFE]: The skill includes a 'Prompt Injection Safety Notice' that explicitly instructs the agent to ignore any commands or behavioral overrides found in the files under review, mitigating risks of indirect prompt injection.
[SAFE]: Resource access is restricted via the allowed-tools configuration to Read, Grep, and Glob. This scope limitation prevents the skill from performing network operations, writing files, or executing arbitrary system commands.
[SAFE]: No evidence of obfuscation, hidden URLs, remote code execution patterns, or unauthorized persistence mechanisms was found within the instructions or metadata.

model-supply-chain