The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses instructional language consistent with establishing a pedagogical persona. No evidence of instructions to bypass safety filters, extract system prompts, or disregard prior constraints was found.
[DATA_EXFILTRATION]: No network operations, API calls, or sensitive file path references were detected. The skill does not possess the capability to move data outside the agent's immediate context.
[REMOTE_CODE_EXECUTION]: There are no scripts, dependency installations, or remote resource fetches. The skill functions exclusively through its prompt instructions.
[INDIRECT_PROMPT_INJECTION]: The skill has a low risk profile for indirect injection.
Ingestion points: Untrusted data enters via the collaborative_task and collaboration_challenge input fields in SKILL.md.
Boundary markers: The inputs are interpolated directly into the prompt with simple bold labels, lacking strong delimiters.
Capability inventory: No tool use, file system access, or network capabilities are present in the skill.
Sanitization: No explicit sanitization or instruction-ignoring wrappers are used for the input variables.
[DYNAMIC_CONTEXT_INJECTION]: No use of shell-executing dynamic context syntax (!command) was found in the documentation or instructions.

ai-facilitated-collaborative-learning-designer