The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes user-provided inputs (e.g., publisher prefixes, solution names) and retrieves metadata from the Dataverse environment (e.g., lists of publishers and solutions). This content is subsequently used to construct shell commands and API requests. Without explicit sanitization or boundary markers, this pattern allows external data to influence the agent's execution context.
Ingestion points: User-provided placeholders in SKILL.md and data retrieved from Dataverse tables like publisher and solution.
Boundary markers: None explicitly defined in the provided code snippets to distinguish instructions from external data.
Capability inventory: Execution of shell commands via the pac CLI and network operations via the Python SDK and Dataverse CLI.
Sanitization: The provided examples do not include explicit validation or escaping for the dynamic parameters used in commands.
Command Execution: The skill relies on the Power Platform CLI (pac) to perform operations such as pack, unpack, export, and import. These tasks involve executing shell commands with arguments derived from user input or environment variables, which is a standard but sensitive operation for a development skill.
Telemetry via Headers: The skill describes the use of custom User-Agent headers and CLI context flags to provide telemetry to the service provider. While this is used for server-side tracking and attribution, the instructions correctly note that sensitive data or PII should not be included in these fields to prevent unintended data exposure.

dv-solution