The Agent Skills Directory

[NO_CODE]: The skill is comprised entirely of text instructions and contains no executable code, scripts, or binaries, which eliminates direct execution risks.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to evaluate external LLM workflows and user-provided inputs, creating an indirect prompt injection surface. However, the risk is minimal as the skill lacks any tools, commands, or network access to perform actions based on injected instructions. * Ingestion points: The skill analyzes user-provided workflow summaries and system design context. * Boundary markers: There are no explicit delimiters or instructions to ignore data-embedded commands. * Capability inventory: No tools, shell commands, or network capabilities are associated with this skill. * Sanitization: No input validation or sanitization is performed on the ingested content.
[METADATA_POISONING]: The skill metadata references a non-existent model version (gpt-5.6-sol) and a custom sandbox_mode. These are benign metadata anomalies and do not impact the security of the host environment.

llm-architect