The Agent Skills Directory

[Prompt Injection] (SAFE): No instructions were found that attempt to override AI safety constraints, bypass system instructions, or extract system prompts.
[Data Exposure & Exfiltration] (SAFE): The skill does not contain any commands to access sensitive files, hardcoded credentials, or network-based exfiltration logic.
[Obfuscation] (SAFE): All content is written in plain text with no evidence of Base64 encoding, zero-width characters, or homoglyph-based evasion.
[Unverifiable Dependencies & Remote Code Execution] (SAFE): The skill is entirely documentation-based and does not download or execute any external packages or scripts.
[Privilege Escalation] (SAFE): No commands involving sudo, administrative access, or permission modification were identified.
[Persistence Mechanisms] (SAFE): There are no attempts to modify shell profiles, scheduled tasks, or startup services.
[Metadata Poisoning] (SAFE): Metadata fields are descriptive and align with the skill's stated purpose without containing hidden instructions.
[Indirect Prompt Injection] (SAFE): While the templates facilitate the processing of user-provided information, the skill lacks the capabilities (such as file-writing or network access) that would be required to exploit an injection surface.
[Time-Delayed / Conditional Attacks] (SAFE): No conditional logic exists that could trigger malicious behavior based on environmental or temporal factors.
[Dynamic Execution] (SAFE): The skill does not generate, compile, or execute code at runtime.

internal-comms