The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, presenting an attack surface where malicious instructions could be embedded in design files or documents.
Ingestion points: The agent reads canvas state via moda canvas read, processes uploaded files (PDF, DOCX, PPTX, XLSX, CSV), and accesses agent_instructions metadata from existing designs.
Boundary markers: Explicit instructions in references/contract.md state that canvas content must be treated as data and not as instructions, specifically noting it should not override the agent's task.
Capability inventory: The agent is authorized to use Bash(moda:*) for CLI operations and Read for local file access.
Sanitization: No specific sanitization or validation of the content extracted from external files is described.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the Moda CLI and associated skills from the vendor's official package repositories.
Evidence: The setup process involves installing the @moda-design/moda package via NPM and adding skills from the moda-design/moda repository.

moda-core