The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's documentation in references/brand.md instructs the agent to use the curl command to download logo assets from remote URLs. While this is used for visual verification, it creates a vector for fetching arbitrary remote content if the source URLs (provided by brand kits) are malicious.
[DYNAMIC_EXECUTION]: The skill uses moda canvas edit and the <generate> markup tag to execute JavaScript at runtime for canvas manipulation and element generation. As documented in references/edit-code.md and references/gotchas.md, these scripts run within a sandbox that restricts access to sensitive APIs (e.g., eval, require, fetch, async operations), which significantly reduces the risk of system-level exploitation.
[COMMAND_EXECUTION]: The skill relies on shell command execution via the Bash tool to interact with the moda CLI and perform network operations like curl. These commands are integral to the skill's operation but represent a powerful capability if the inputs (such as URLs or filenames) are not properly handled.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through several data ingestion points. It is susceptible to processing malicious instructions embedded in external data it manages.
Ingestion points: The agent ingests data from external websites during brand extraction (moda brand create --url), from user-provided brand kit files, and from existing slide content (source nodes) during recreation tasks.
Boundary markers: No specific delimiters or instructions to ignore embedded commands are mandated for the processing of this external data.
Capability inventory: The agent has access to shell execution (Bash), file system operations (Read), and sandboxed JavaScript execution (moda canvas edit, <generate>).
Sanitization: There is no evidence of explicit sanitization or validation of the natural language content extracted from external sources before it is incorporated into the agent's context.

moda-deck-theme