The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of review_scope and context_summary which are interpolated directly into task prompts for expert sub-agents in Phase 1. This creates a potential surface where malicious instructions embedded in the material being reviewed could attempt to influence the sub-agents' logic.
Ingestion points: debate_input fields review_scope and context_summary in SKILL.md.
Boundary markers: Absent; inputs are directly interpolated into the template strings.
Capability inventory: The skill utilizes Read, Write, Task, and restricted Bash(mkdir *) tools.
Sanitization: No specific sanitization or escaping of the user-provided scope is performed before prompt construction.

debate-protocol