The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it is designed to process untrusted user input (opponent's debate arguments) and has access to sensitive tools such as run_terminal_cmd and read_file without explicit boundary markers or sanitization.
Ingestion points: The skill is intended to process external arguments during debates (SKILL.md).
Boundary markers: Absent.
Capability inventory: run_terminal_cmd, read_file, grep (SKILL.md).
Sanitization: Absent.
[SAFE]: The skill contains multiple self-referential claims regarding its own accuracy and completeness (e.g., 'Research Validation Score: 97/100', 'Technical Accuracy: 100%'). These claims are evaluated as metadata and not treated as authoritative conclusions about the skill's security.
[SAFE]: The skill references various well-known AI services and tools (e.g., Anthropic, OpenAI, GitHub Copilot) as part of its debate points, which is consistent with its stated educational purpose.

ai-debate-dominator