The Agent Skills Directory

[SAFE]: No security vulnerabilities were detected. The skill functions as a template for prompt generation and does not execute arbitrary code, access sensitive files, or perform unauthorized network operations.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided data (prop name and description) which is interpolated into a prompt. While this creates a theoretical surface for indirect prompt injection, the skill includes robust boundary instructions—such as strictly prohibiting narrative elements, characters, and specific keywords—which act as internal guardrails for the agent's output.
Ingestion points: User-provided name and description variables are used in SKILL.md.
Boundary markers: Specific "Forbidden items" and "Generation rules" are used to constrain the model's focus.
Capability inventory: Uses the save_prop_final_prompt tool to store output.
Sanitization: The prompt instructs the agent to exclude narrative and character-based descriptions, effectively filtering the input during the generation process.

prop-prompt