The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its input schema.
Ingestion points: Untrusted data enters the prompt via the original_task, learner_profile, learning_objective, student_profiles, and available_support fields in SKILL.md.
Boundary markers: The instructions do not employ delimiters (such as XML tags or triple quotes) to separate the teacher-provided content from the system instructions, which could allow an attacker to override the agent's behavior.
Capability inventory: The skill does not define any external tools, scripts, or network capabilities, which limits the potential impact of a prompt injection to the manipulation of the generated text output.
Sanitization: There is no evidence of validation or sanitization logic to prevent malicious instructions embedded in the input fields from influencing the model's logic.

differentiation-adapter