The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill possesses a significant vulnerability surface due to its core functionality.
Ingestion Points: Untrusted data enters the agent context via the integrate_external action (which implies processing external API specifications) and through the general task descriptions provided by users or external triggers as seen in references/GUIDE.md.
Boundary Markers: There are no explicit delimiters or 'ignore embedded instructions' warnings defined in SKILL.md or the associated scripts to prevent the LLM from obeying instructions embedded within the data it is processing.
Capability Inventory: The skill has powerful 'write' capabilities, as it generates executable backend logic, API routes, and authentication implementations (code and files properties in SKILL.md).
Sanitization: There is no evidence of content validation, escaping, or sanitization for external content before it is interpolated into the code generation process.
Dynamic Execution (LOW): While SKILL.md contains logging hooks (on_invoke, on_success) that appear to be template strings, the risk is minimized because the input parameters (action, runtime) are strictly constrained by enums in the parameter schema.

backend-development