The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection due to its handling of untrusted external data.
Ingestion points: Untrusted data enters the agent context through moda brand create --url (website extraction), moda template pull (team templates), and by reading existing canvas nodes as described in SKILL.md and references/brand.md.
Boundary markers: The instructions do not specify the use of delimiters or warnings to ignore embedded instructions when processing or re-inserting external content.
Capability inventory: The skill utilizes the Bash(moda:*) tool for powerful operations including moda canvas markup, moda canvas edit, and moda export.
Sanitization: There are no instructions for sanitizing or validating external content before it is interpolated into prompts or executed via markup/edit tools.
[DYNAMIC_EXECUTION]: The skill creates and executes JavaScript logic at runtime to modify the document canvas.
Evidence: references/edit-code.md and references/markup.md detail the use of moda canvas edit and the <generate> tag, which allow the agent to write and run JavaScript for batch mutations and element generation. While described as sandboxed, the documentation notes the sandbox is not a hard security boundary.
[COMMAND_EXECUTION]: The skill uses the Bash tool to invoke the moda CLI suite for document operations.
Evidence: SKILL.md and several reference files (e.g., references/brand.md) contain commands for moda canvas, moda brand, moda media, and moda export to manage the document lifecycle.
[EXTERNAL_DOWNLOADS]: The skill performs remote downloads of image assets for inspection.
Evidence: references/brand.md and references/templates.md instruct the agent to use curl -o /tmp/... "<url>" to download logo variants and template thumbnails. The URLs are short-lived, signed preview links generated by the platform's metadata tools.

moda-document-print