The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by navigating to arbitrary URLs and extracting DOM structure and CSS properties.
Ingestion points: TARGET_URL (SKILL.md) derived from user arguments.
Boundary markers: Absent. The instructions do not provide delimiters or warnings to ignore embedded instructions within the scraped content.
Capability inventory: The skill utilizes Bash, Write, and Skill("pinchtab"), allowing for filesystem modifications and further browser interactions.
Sanitization: Absent. The extracted data is written directly into an output report file without filtering or escaping.
[COMMAND_EXECUTION]: The skill uses the Bash tool for local environment setup.
Evidence: Execution of mkdir -p to create the output directory for screenshots.
[CREDENTIALS_UNSAFE]: The skill instructions direct the agent to "attempt to find credentials from environment variables" if a page requires authentication. While functional, this encourages the agent to search for sensitive data in its environment, which could lead to unintended exposure if the environment contains secrets unrelated to the target site.

page-inspector