The Agent Skills Directory

Command Execution: The skill utilizes the Bash tool to execute internal Node.js scripts located in the skill's root directory. These scripts, such as resolve-environment.js and dataverse-request.js, are used to read local configuration files and perform authenticated Dataverse Web API operations. This is an expected pattern for managing cloud environment resources.
Indirect Prompt Injection Surface: The skill ingests untrusted data in the form of User Principal Names (UPNs) and team names via command-line arguments or interactive prompts. These values are interpolated into OData filter strings for Dataverse API requests (e.g., domainname eq '<upn>').
Ingestion points: Ingests data through $ARGUMENTS (--user, --team) and AskUserQuestion responses.
Boundary markers: The skill does not explicitly define delimiters for these inputs within the instruction set.
Capability inventory: Uses the Bash tool to execute GET, POST, and DELETE operations via dataverse-request.js.
Sanitization: The skill instructions do not explicitly detail sanitization or escaping logic for these interpolated values, which could theoretically allow for OData query injection. However, the risk is mitigated by a mandatory confirmation gate in Step 4 where the agent must present the resolved entities to the user for approval before execution.

assign-offline-profile