The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to fetch and process data from external Dataverse and ERP environments. This creates a surface where external data could potentially contain instructions that influence the agent's behavior.
Ingestion points: Data enters the agent's context through client.records.list(), client.query.sql(), client.query.fetchxml(), and dataverse data query calls in SKILL.md and references/erp-reads.md.
Boundary markers: The provided code samples do not include specific delimiters or instructions for the agent to ignore embedded commands within the retrieved data.
Capability inventory: Across the skill files, the agent has capabilities including command execution via the dataverse CLI, network operations via the PowerPlatform-Dataverse-Client SDK and urllib, and local file writing (e.g., to_csv in references/querybuilder.md).
Sanitization: The skill documentation does not demonstrate explicit sanitization or filtering of the retrieved data content before it is processed.
Command Execution: The skill provides instructions for driving the dataverse CLI tool directly for data queries and counts. This involves the agent executing external shell commands, which is the intended mechanism for performing headless Dataverse operations.
External Package Dependencies: The instructions for Jupyter notebook setup in references/jupyter-setup.md recommend the installation of several Python packages, including PowerPlatform-Dataverse-Client and azure-identity. These are recognized libraries from a well-known organization, used for their intended purpose of providing Dataverse connectivity and authentication.
Network Operations: The skill performs network requests to interact with Dataverse and ERP APIs. These operations, documented in references/web-api-advanced.md using urllib, target the environment URL specified in the system's environment variables and include necessary authentication headers.

dv-query