The Agent Skills Directory

Command Execution: The skill uses the Power Platform CLI (PAC) and a Python-based SDK to perform administrative tasks like updating organization settings and managing bulk delete jobs. These operations are essential for the skill's intended purpose of environment administration.
Safety Gates for Destructive Actions: For operations that are irreversible, such as bulk deletes, the skill implements strict confirmation rules. It requires the agent to verify that the user explicitly wants to delete all records and identifies the specific entity logical name before executing any commands.
Indirect Prompt Injection Surface: The skill allows for the ingestion of user-provided FetchXML filters and entity names. This represents a potential surface where malicious instructions could be embedded in data; however, the skill mitigates this by requiring prose previews and explicit human-in-the-loop confirmations for stateful changes.
Ingestion points: User-provided FetchXML strings and entity logical names in SKILL.md.
Boundary markers: The skill explicitly instructs the agent to ask for confirmation and use specific keyword gates like 'ALL' for destructive actions.
Capability inventory: Shell execution of pac CLI commands and Python-based Web API CRUD operations across all scripts.
Sanitization: The skill relies on manual user confirmation and descriptive previews rather than automated data sanitization.
Vendor-Specific Tooling: The skill includes instructions to ensure the latest version of the PAC CLI is installed via pac install latest. As these tools originate from the vendor's own infrastructure, this is considered a standard maintenance and update procedure.

dv-admin