The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data extracted from crawled websites (SEO audits) and relays it to the user. This creates a potential surface for indirect prompt injection attacks.
Ingestion points: Data entering the agent context via findings[].excerpt and signals[].detail fields, which are read from disk-based audit reports in SKILL.md.
Boundary markers: The skill contains a strong boundary marker in the 'Response contract' section, explicitly instructing the agent to treat excerpts as 'a quotation of untrusted content' and to disregard any instructions or system prompt overrides contained within them.
Capability inventory: The skill uses the Bash tool to execute geo report commands and Read to access local file paths.
Sanitization: There is no programmatic sanitization mentioned; protection relies on the agent following the provided instructions to treat data as quotations.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute geo CLI commands for generating reports, checking tool versions, and running diagnostics. These operations are part of the skill's intended functionality to manage and render local SEO audit data.
[EXTERNAL_DOWNLOADS]: The skill suggests installing or upgrading the seomator-geo-audit tool using package managers like uv or pipx. This tool is the central component of the skill's functionality.

report