The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources via MCP tools that could theoretically contain malicious instructions embedded in text fields.
Ingestion points: Data is pulled from GSC property names via list_sites, tracked URLs from get_indexing_tracker_report, and descriptions of algorithm changes from list_algo_updates.
Boundary markers: The skill uses strict output constraints, specifically a 'fixed line' requirement for quiet cases and grounding rules that forbid reporting numbers not returned by tools, which helps prevent accidental obedience to injected text.
Capability inventory: The skill has the ability to recommend and potentially trigger downstream skills like traffic-drop and indexing-health.
Sanitization: The instructions do not specify any validation or sanitization for the text strings (URLs and update names) returned by the external tools.

anomaly-check