The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, including client-provided offer text and scraped brand websites, which could contain malicious instructions designed to influence the agent's HTML generation process.
Ingestion points: Data from offer.txt and the brand-url website.
Boundary markers: The agent is instructed to visually verify the output, but there is no technical enforcement of boundaries or sanitization of the untrusted data before it enters the context.
Capability inventory: Access to Bash for command execution and Write for file creation.
Sanitization: None observed in the provided logic.
[COMMAND_EXECUTION]: The skill invokes shell commands via the Bash tool to execute Python scripts and launch headless Chrome for creating brand screenshots and PDFs.
[EXTERNAL_DOWNLOADS]: The skill connects to arbitrary brand websites and product catalogue URLs to download styling metadata and item imagery.
[DYNAMIC_EXECUTION]: The scripts/run.py script utilizes os.execv() to re-execute the application within a virtual environment, which involves dynamic process replacement.

proposal-maker