The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill accepts several user-controlled input fields, including skill_to_teach, student_level, and student_profiles, which are interpolated directly into the instructions. This creates a surface for indirect prompt injection, as malicious text provided by a user could theoretically attempt to steer the agent's behavior. However, the potential impact is constrained because the skill lacks dangerous capabilities such as tool execution, network access, or file system modifications.
Ingestion points: User inputs are interpolated in the prompt section of SKILL.md using variables like {{skill_to_teach}} and {{student_profiles}}.
Boundary markers: Absent; the inputs are placed directly within the instructional text without delimiters or isolation instructions.
Capability inventory: No tool usage, file system access, or network requests are defined in the skill.
Sanitization: No input validation or sanitization logic is implemented.

explicit-instruction-sequence-builder