The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external websites to generate SEO reports, creating a surface for indirect prompt injection attacks.
Ingestion points: External HTML content is fetched and parsed by scripts/fetch_page.py and scripts/parse_html.py.
Boundary markers: The skill does not implement explicit delimiters or instruction isolation for the fetched content before processing by the LLM.
Capability inventory: The skill can execute local Python scripts, perform network requests, and write screenshot files to the disk.
Sanitization: While scripts extract specific HTML elements, the text content is used for content quality analysis without specialized sanitization against prompt injection.
[COMMAND_EXECUTION]: The skill relies on several local Python scripts to perform visual analysis and webpage fetching.
Evidence: scripts/analyze_visual.py, scripts/capture_screenshot.py, and scripts/fetch_page.py are executed to gather data.
Mitigation: These scripts implement robust security checks, including _is_public_host to prevent Server-Side Request Forgery (SSRF) and path validation to prevent directory traversal when saving screenshots.
[EXTERNAL_DOWNLOADS]: The skill documentation references the use of external tools for performance auditing.
Evidence: references/cwv-thresholds.md mentions the use of npx lighthouse and curl for interacting with the PageSpeed Insights API, which are well-known developer tools.

seo