The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation process requires cloning an external repository (https://github.com/ucsandman/marketing-studio.git) and executing scripts from it, such as 'node scripts/install-skills.mjs' and 'python launch.py'. This enables the execution of arbitrary code from a third-party source not associated with the vendor.
[EXTERNAL_DOWNLOADS]: The skill relies on fetching its core engine, feeder scripts, and configuration builders from a non-whitelisted GitHub repository.
[DYNAMIC_EXECUTION]: The 'getBrand' function in 'studio/src/lib/brand.ts' uses 'require()' with a template string derived from user-supplied input: 'require(../../brands/${brandId}.json)'. This pattern is vulnerable to directory traversal if the brand ID is not strictly sanitized, potentially allowing the agent to load and expose other JSON files on the filesystem.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources and local files to generate marketing assets, creating a surface for injection.
Ingestion points: Brand data is loaded from 'brands/.json' and product demos are captured from external URLs via Playwright.
Boundary markers: No explicit delimiters or boundary instructions are present to prevent the agent from following instructions embedded in the design system or the captured web content.
Capability inventory: The skill has the capability to execute shell commands via 'npx', 'node', and 'python', as well as perform network operations to ElevenLabs.
Sanitization: While a Zod schema is used to validate the JSON structure, it does not filter for malicious instructional content within the strings.
[COMMAND_EXECUTION]: The skill frequently invokes shell commands including 'git', 'npm', 'npx', 'node', and 'python' to manage the rendering pipeline and toolchain.

marketing-studio-agent