The Agent Skills Directory

Indirect Prompt Injection Surface: The skill's diagnostic workflows (detailed in references/operations/pipeline-diagnosis.md and references/operations/automated-diagnostic-workflow.md) ingest and process external data, such as Spark driver/executor logs and pipeline activity runs. This data, which originates from the Spark runtime and may be influenced by external inputs, represents an ingestion surface where indirect instructions could be embedded.
Ingestion points: queryActivityRuns output, driver/executor logs, and notebook cell contents.
Boundary markers: The instructions do not specify explicit delimiters or warnings to ignore embedded instructions in the processed data.
Capability inventory: The skill utilizes az rest for resource management (including POST, PATCH, and DELETE operations) and curl for data retrieval.
Sanitization: No explicit sanitization or filtering of external log content is mentioned before processing.
Command Execution via CLI Tools: The skill makes extensive use of az rest, curl, and wget to interact with Microsoft Fabric APIs and manage resources across different modes (e.g., SKILL.md, references/authoring/resources/notebook-api-operations.md). While these capabilities are essential for the skill's purpose as a developer tool, they represent powerful system-level interactions.
External Data Access: The skill facilitates the download of Spark event logs from OneLake and references the setup of a local Spark History Server (references/operations/spark-history-server.md). These operations target official Microsoft Fabric endpoints (onelake.dfs.fabric.microsoft.com) and are part of the documented diagnostic procedures.

spark-cli