The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze untrusted data from URLs and code repositories, which serves as an ingestion point for potential indirect prompt injections.
Ingestion points: SKILL.md workflow instructions for URLs and repository apps.
Boundary markers: None defined in instructions to delimit external content.
Capability inventory: Access to file system and instruction to run applications.
Sanitization: No mention of content sanitization.
[DYNAMIC_EXECUTION]: The workflow instructions in SKILL.md encourage the execution of the target application code ('run the app when feasible') to support the design audit process.

design-audit