The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a CLI tool named seomator-geo-audit via the geo command. It includes standard preflight checks to ensure version compatibility (e.g., geo --version) and provides instructions for the agent to install or upgrade the tool using uv or pipx if missing.
[INDIRECT_PROMPT_INJECTION]: The skill explicitly defines a security boundary for data ingested during the crawl. The instructions in SKILL.md mandate that the agent treat findings[].excerpt and signals[].detail (which contain content from external websites) as untrusted quotations. It provides clear instructions to report any instruction-like patterns found in these excerpts as site findings rather than executing them.
[SAFE]: The skill uses allowed-tools to limit the agent's environment and provides a 'Response Contract' that prevents the agent from inventing data or performing unauthorized arithmetic, ensuring output fidelity.

audit