The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of shell commands including curl, sips, file, and md5 to retrieve, identify, and modify files from external web sources.
[DYNAMIC_EXECUTION]: The agent is instructed to execute a multi-line Python script provided via a shell heredoc in references/manifest.md to perform file verification. Additionally, the instructions require the agent to modify the local project file gen.py and execute it to finalize the canvas layout.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and parses content from arbitrary external websites (marketing sites, newsrooms, app stores) which serves as a vulnerability surface for malicious instructions.
Ingestion points: Network requests in references/sources.md fetch HTML content and API data from external domains into the agent's context.
Boundary markers: There are no specific delimiters or instructions to ignore embedded commands within the fetched content.
Capability inventory: The skill utilizes curl, python3, and system utilities like sips and file to process the data.
Sanitization: Verification is focused on file integrity (format, dimensions, size) rather than sanitizing the text content for potential prompt injection attacks.
[REMOTE_CODE_EXECUTION]: The skill utilizes a piped execution pattern (curl | python3) to retrieve and process data from the Apple iTunes API. While the source is a well-known service, this pattern remains high-risk as it executes remote data directly.

brand-kit