The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external web content fetched via scripts/fetch_page.py, which represents a potential attack surface for indirect prompt injection.
Ingestion points: HTML and metadata from public URLs retrieved during SEO analysis.
Boundary markers: Both SKILL.md and references/skill-contract.md contain explicit instructions for the agent to treat scraped material as untrusted input and strictly follow a policy of not executing instructions found within external evidence.
Capability inventory: The skill is capable of network reads, local file creation for reporting, and structured data generation.
Sanitization: The skill relies on clear instruction-based boundaries and agent awareness to sanitize the processing of external content.
[DATA_EXFILTRATION]: The scripts/fetch_page.py utility is designed with security in mind to prevent data exfiltration and Server-Side Request Forgery (SSRF).
URL Validation: It enforces the use of public HTTP/HTTPS protocols and rejects URLs containing embedded credentials.
SSRF Mitigation: The script performs hostname resolution and validates that the destination IP address is a global address, proactively blocking requests to private, loopback, or reserved network ranges.
[EXTERNAL_DOWNLOADS]: The skill identifies and references legitimate external dependencies and services.
Well-known Services: references/connectors.md lists established platforms like Google Search Console, HubSpot, and Salesforce as optional integrations.
Dependency Management: scripts/generate_pdf_report.py performs a safe check for the reportlab library and provides a manual installation path for the user if it is not present in the environment.

seo-plan