The Agent Skills Directory

[Prompt Injection] (SAFE): No attempts to override agent behavior or bypass safety guidelines were detected. The instructions focus on establishing a specialized persona and grounding responses in provided documentation.
[Data Exposure & Exfiltration] (SAFE): No patterns for accessing sensitive files (e.g., .env, .ssh) or exfiltrating data via network requests (e.g., curl, fetch) are present.
[Obfuscation] (SAFE): The skill contains no Base64, zero-width characters, homoglyphs, or other encoding techniques intended to hide malicious content.
[Unverifiable Dependencies & RCE] (SAFE): There are no package installations (npm/pip) or instructions to download and execute remote scripts.
[Indirect Prompt Injection] (SAFE): The skill processes untrusted user data (smart contract code); however, it lacks the capabilities (network, file-system, or code execution) to be exploited. It includes grounding instructions to maintain its persona.
[Dynamic Execution] (SAFE): No use of dynamic code execution (eval/exec) or runtime compilation was found.

defi-architect