The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by fetching web pages through the scripts/fetch_page.py utility, which creates a potential surface for indirect prompt injection attacks.
Ingestion points: The utility scripts/fetch_page.py retrieves HTML content from external URLs for SEO metadata extraction.
Boundary markers: The SKILL.md and references/skill-contract.md files contain explicit instructions to treat all scraped page text and external evidence as untrusted input and to disregard any instructions embedded within that data.
Capability inventory: The skill environment possesses network read capabilities (via urllib.request) and report generation capabilities (Markdown and PDF). It is instructed not to perform persistent writes or external mutations without explicit user authorization.
Sanitization: The fetch_page.py script utilizes regular expressions for targeted metadata extraction rather than rendering or executing the external content, and it includes robust SSRF protections by validating URLs and blocking private/reserved IP addresses.

seo-content