The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external websites via the scripts/fetch_page.py utility. This creates a potential attack surface where malicious instructions could be embedded in the HTML of a site being audited.
Ingestion points: External web content is fetched and parsed in scripts/fetch_page.py.
Boundary markers: The skill contains multiple strong mitigations. SKILL.md explicitly instructs the agent to treat scraped text as untrusted and not to follow embedded instructions. references/skill-contract.md reinforces this as a core invariant.
Capability inventory: The skill utilizes network read capabilities (scripts/fetch_page.py) and local file writing for report generation (scripts/generate_markdown_report.py, scripts/generate_pdf_report.py, scripts/schema_generator.py).
Sanitization: scripts/fetch_page.py performs specific metadata extraction (title, description, h1 count) using regular expressions rather than executing or broadly interpreting the content. scripts/schema_generator.py uses proper JSON escaping for injected values.
[COMMAND_EXECUTION]: The skill includes local Python scripts for data processing and reporting. These scripts follow security best practices:
scripts/fetch_page.py implements thorough URL validation, blocking private, loopback, and reserved IP addresses to prevent SSRF attacks. It also explicitly blocks URLs containing embedded credentials.
scripts/generate_pdf_report.py references the well-known reportlab library for PDF generation but does not attempt to install it automatically, instead providing a clear error message to the user.

seo-google