The Agent Skills Directory

[COMMAND_EXECUTION]: The file references/design-picker.md contains instructions for the agent to execute shell commands to start a local web server (python3 -m http.server 8723 &). This operation exposes the local project directory over the network, creating a potential attack surface if the environment is not isolated.
[DYNAMIC_EXECUTION]: The skill's core scripts (scripts/package-loader.mjs and scripts/contrast-report.mjs) are described as having the ability to 'bootstrap' the HyperFrames package. This involves dynamically loading and potentially installing external code at runtime, which can bypass static security checks.
[COMMAND_EXECUTION]: Instructions in references/typography.md direct the agent to write a Python script to a temporary path (/tmp/fontquery.py) and execute it. Running dynamically generated scripts from temporary locations is a high-risk pattern that can be exploited for arbitrary command execution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and parse third-party design specifications from files like frame.md or design.md. These inputs are then interpolated into HTML and CSS templates for video rendering.
Ingestion points: frame.md, design.md, and DESIGN.md in the project root.
Boundary markers: The skill does not define explicit delimiters or warnings to ignore embedded instructions in these files.
Capability inventory: The skill utilizes python3 for data extraction and http.server for internal tooling, alongside GSAP for animation generation.
Sanitization: There is no evidence of sanitization or validation of the content parsed from external design specs before it is processed by the agent or rendered into templates. (Severity: LOW)

hyperframes-creative