The Agent Skills Directory

[Prompt Injection] (SAFE): The instructions are strictly focused on code analysis patterns. There are no attempts to override system prompts or bypass safety filters.
[Data Exposure & Exfiltration] (SAFE): No commands are present that access sensitive local files (like SSH keys) or transmit data to external domains. The network URLs mentioned (AWS/GCP/Azure metadata) are provided as examples of what to detect in PHP code, not for the agent to access.
[Obfuscation] (SAFE): No encoded content, zero-width characters, or homoglyphs were detected in the skill file.
[Unverifiable Dependencies & RCE] (SAFE): The skill does not define or install any external packages. It provides search patterns (Grep) for static analysis but does not execute the code it analyzes.
[Indirect Prompt Injection] (LOW): The skill is designed to process untrusted PHP code from external sources. While this creates an ingestion surface, the skill's capabilities are limited to search and display (static analysis). It lacks write or execution privileges that would escalate this to a higher risk level.
[Dynamic Execution] (SAFE): No runtime code generation or unsafe deserialization patterns are present. The shell_exec example is part of a code snippet intended for detection, not execution by the agent.

acc-check-ssrf