The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from external websites, which creates an attack surface for indirect prompt injection.
Ingestion points: The scripts/inspect_site.py script extracts the document.title and CSS variable values from external websites and saves them into a report.md file that the agent is instructed to read.
Boundary markers: The extracted content is placed in a Markdown report, but no explicit delimiters or instructions are provided to the agent to ignore potential commands embedded in the website's metadata or styles.
Capability inventory: The agent has access to Bash, Read, and Write tools, which could be leveraged if the agent follows malicious instructions found in the processed content.
Sanitization: The script implements a check to prevent the downloading of local files (via file:// protocols) unless the base URL being inspected is also local. However, it does not sanitize the text content (like titles) against prompt injection patterns.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the playwright Python package and the downloading of Chromium browser binaries to function.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute its internal Python inspection script and perform file operations such as cropping screenshots using the PIL library.

website-visual-inspector