The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external source material and user-provided inputs to generate artifacts, creating a surface where untrusted data could potentially influence agent behavior or tool parameters.
Ingestion points: The workflow involves inspecting and processing existing files, inputs, and user source material as defined in SKILL.md.
Boundary markers: There are no instructions providing specific delimiters or "ignore embedded instructions" warnings when handling user-supplied content.
Capability inventory: The skill utilizes the sandbase_run tool to execute external capabilities and API calls based on processed data (references/sandbase-api-map.md).
Sanitization: The instructions do not specify any validation, filtering, or escaping mechanisms for external content before it is used in the workflow.

hyperframes-keyframes