The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources that could contain malicious instructions designed to influence the agent's analysis or recommendations.
Ingestion points: Untrusted strings enter the context via list_experiments (experiment names and descriptions) and list_annotations (label and description fields from site logs).
Boundary markers: The instructions do not define clear delimiters or "ignore instructions" wrappers for the data returned by the tools before processing them into a recommendation.
Capability inventory: The skill has the capability to write data back to the environment via the create_annotation tool and generates high-level project recommendations (roll out, extend, or stop).
Sanitization: No explicit sanitization or validation of the text content from the Search Console or experiment metadata is mentioned.

experiment-results