The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted design artifacts and build records, creating a surface for indirect prompt injection.\n- Ingestion points: The skill reads the design/ directory and various build artifacts to perform its audit (SKILL.md).\n- Boundary markers: There are no explicit instructions, such as delimiters or system prompts, directing the agent to ignore or isolate instructions found within the design artifacts.\n- Capability inventory: The skill is instructed to write analysis results and "debt" items to validation.md and decisions.md (SKILL.md).\n- Sanitization: No input validation or sanitization mechanisms are specified to prevent adversarial content in the design records from influencing the agent's reporting or file-writing tasks.

product-review