The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes data from external Power Automate Desktop flows, which represents a potential surface for indirect prompt injection if the flow content is adversarial.
Ingestion points: The agent retrieves potentially untrusted flow names, statuses, and run details via tools like mcp__flowagent__list_desktop_flows and mcp__flowagent__get_run_details as described in SKILL.md.
Boundary markers: The instructions do not specify explicit delimiters or "ignore embedded instructions" warnings for the agent when interpreting flow data.
Capability inventory: The skill allows the agent to perform significant actions including mcp__flowagent__run_desktop_flow, mcp__flowagent__delete_flow, and general command execution via Bash.
Sanitization: There are no defined mechanisms to sanitize or filter metadata or output retrieved from the RPA environment before it is processed by the agent.

manage-desktop-flows