The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to facilitate the querying and processing of telemetry data (such as error messages and stack traces) emitted by client-side applications. As this data originates from external sources, it represents a potential surface for indirect prompt injection if an attacker-controlled telemetry event contains malicious instructions aimed at the analyzing agent.
Ingestion Points: Telemetry data is ingested through the kusto_query tool when accessing tables such as Office_Fluid_FluidRuntime_Error and Office_Fluid_Scriptor_Error (referenced in SKILL.md and references/kusto-query-reference.md).
Boundary Markers: The skill does not explicitly define prompt-level delimiters, though it includes instructions for users to filter specific noise sources (e.g., reverse proxy traffic) to maintain data integrity.
Capability Inventory: The skill enables the agent to generate and potentially execute Kusto queries using the kusto_query tool, which is necessary for data analysis.
Sanitization: The instructions include best practices for filtering known noisy or untrusted domains from telemetry (e.g., .goskope.com, .myshn.net), which serves as a form of data validation.
Internal Configuration Disclosure: The skill documents internal cluster URLs (kusto.aria.microsoft.com) and database GUIDs. As the skill is authored by the vendor (Microsoft) for use within its own infrastructure, these references are considered standard operational configuration and are managed via VPN access and internal security groups (e.g., M365HeartbeatTenantUsers).

ff-oce-kusto