The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to read local files from the user's home directory.
Evidence: ~/Desktop/Canva_LP_Templates/ is referenced in SKILL.md and examples/README.md as a required resource for design analysis.
This instruction grants the agent access to the local file system on the user's Desktop.
[INDIRECT_PROMPT_INJECTION]: The skill populates HTML templates with untrusted user input, which could lead to malicious code execution in the generated landing page.
Ingestion points: User input is used to fill placeholders in references/SECTIONS.md, such as {{店名}} (Store Name) and {{オファー価格}} (Offer Price).
Boundary markers: The skill does not provide clear delimiters or instructions to prevent the agent from interpreting instructions embedded within the user-provided data.
Capability inventory: The agent can read local files and generate executable web code (HTML/CSS).
Sanitization: The skill lacks any input validation or sanitization mechanisms to escape special characters, making the generated pages vulnerable to XSS.
[EXTERNAL_DOWNLOADS]: The skill utilizes assets from a well-known third-party service.
Evidence: Typography settings in references/TYPOGRAPHY.md and references/SECTIONS.md link to Google Fonts (fonts.googleapis.com).

jp-store-lp