The Agent Skills Directory

Prompt Injection (SAFE): No malicious prompt injection or behavior override patterns detected. The skill contains examples of prompt instructions, but these are clearly marked as documentation or "anti-patterns" to avoid.
Data Exposure & Exfiltration (SAFE): No hardcoded credentials, sensitive file paths, or unauthorized network operations. Use of tools is limited to read-only operations (Read, Grep, Glob) for documentation lookup.
Obfuscation (SAFE): No evidence of Base64, zero-width characters, homoglyphs, or other obfuscation techniques.
Unverifiable Dependencies & Remote Code Execution (SAFE): The skill does not perform any package installations or remote code downloads. Links provided in the documentation point to legitimate platforms (Anthropic, OpenAI, Vercel) and are for reference purposes only.
Indirect Prompt Injection (SAFE): While the skill ingests data from its own reference files, it does not interact with untrusted external data sources in a way that creates a high-risk injection surface. The capabilities are restricted to read-only tools.
Metadata Poisoning (SAFE): Metadata is accurate and descriptive. While the documentation references future-dated or fictional AI models (e.g., Claude 4.5, GPT-5), this appears to be illustrative or forward-looking and does not present a security risk.

agent-prompt