The Agent Skills Directory

[SAFE]: The skill performs read-only analysis of public web content. It includes explicit safety boundaries such as respecting robots.txt directives, enforcing a 50-page crawl limit, and implementing rate limiting (1-second delay between fetches) to prevent accidental Denial of Service (DoS) behavior.
[SAFE]: All network operations are directed at the target audit URL and well-known search engine/AI crawler metadata files (e.g., llms.txt, robots.txt). There are no signs of unauthorized data exfiltration or hardcoded credentials.
[SAFE]: The skill uses specialized subagents for analysis, but the primary logic is instructional and descriptive, focusing on SEO/GEO heuristics rather than executable code injection or privilege escalation.
[EXTERNAL_DOWNLOADS]: The skill references a community GitHub repository (zubair-trabzada/geo-seo-claude) for documentation and upstream context. These references are documented neutrally and follow best practices for attribution without executing remote scripts.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from target websites (headings, meta tags, schema). However, it defines strict processing boundaries and classification rules, and its output is a structured markdown report, which minimizes the risk of indirect injection influencing the host agent's core instructions.

geo-audit