stage-dsl
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a document model that includes fields for agent instructions and learner orientations, such as
systemPromptandlearnerBriefwithin the Project-Based Learning (PBL) project structure. Because thepatch_stagetool allows writing arbitrary values to these paths, it creates a surface where malicious instructions could be embedded in the document. - Ingestion points: The
patch_stagetool is used to update fields inSKILL.mdandreferences/pbl.md. - Capability inventory: The toolset includes
patch_stagefor writing andread_stagefor reading scene content. - Boundary markers: There are no documented delimiters or instructions to treat user-supplied input as untrusted when patching these fields.
- Sanitization: The documentation does not specify sanitization for the instructional fields, though it mentions rendering-level sanitization for visual styles.
- [DYNAMIC_EXECUTION]: The
interactivewidget content model allows for the storage and execution of complete HTML documents within thehtmlfield, which are rendered via iframesrcDoc. While this is a core functionality for the intended course platform, it provides a mechanism for dynamic code execution within the learner's context.
Audit Metadata