The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by ingesting untrusted data from external PowerPoint files.\n- Ingestion points: The skill imports files using moda canvas import-pptx and moda file upload as described in SKILL.md and references/markup.md.\n- Boundary markers: The instructions lack explicit delimiters or warnings to ignore instructions embedded within the imported slide content. In fact, references/design-quality.md mandates verbatim preservation of source text.\n- Capability inventory: The agent has access to the moda CLI, which allows for file system writes (moda export), network operations (moda file upload, moda media *), and dynamic code execution (moda canvas edit).\n- Sanitization: No sanitization or filtering of the imported slide text is mentioned before it is processed by the agent.\n- [DYNAMIC_EXECUTION]: The skill relies on runtime JavaScript execution to perform its primary tasks.\n- As documented in references/edit-code.md, the moda canvas edit command executes JavaScript files (e.g., fix.js) generated by the agent to mutate the canvas state.\n- references/markup.md describes the <generate> element, which executes a JavaScript block to emit elements at runtime.\n- While these are core functionalities of the moda ecosystem, they represent dynamic execution of agent-authored code.

moda-deck-pptx