The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to create a local shell script (scripts/fetch-stitch.sh) and execute it using bash to download assets.
[REMOTE_CODE_EXECUTION]: The agent is directed to generate a shell script at runtime and invoke it with dynamic parameters (download URLs and file paths). This pattern of script generation and execution with external input presents a potential command injection surface if the variables are not rigorously sanitized.
[EXTERNAL_DOWNLOADS]: The skill automates the retrieval of HTML and image assets from external URLs provided by the Stitch MCP tools.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via the generated local gallery.
Ingestion points: Project titles and screen metadata retrieved from the external Stitch API via mcp__stitch__list_screens and mcp__stitch__get_screen tools.
Boundary markers: No boundary markers or 'ignore' instructions are provided for the generated HTML content.
Capability inventory: The skill has file system write access (Write tool) and shell execution capabilities (Bash tool).
Sanitization: While the skill instructs the agent to strip special characters for file paths (slugs), it lacks explicit sanitization instructions for metadata interpolated into the gallery's display text, posing a risk of Cross-Site Scripting (XSS) when the gallery is opened in a browser.

stitch-download