AI Agent Launch Brief Generator

Workflow

Workflow label Highest side effect User-facing surface

Data touched

Private docs or uploaded files Support tickets or CRM notes Customer-visible output Memory, retrieval, or saved state Browser pages or MCP server context

Tool actions

Read from external tools Write to files, tickets, docs, or databases Send messages or post publicly Delete, deploy, purchase, or run code Call MCP tools across more than one server

Controls already in place

Human approval for side effects Scoped credentials and minimal permissions Non-sensitive logs and receipts Rollback or disable switch Red-team fixtures for prompt injection

Safety boundary

Do not paste live credentials, access tokens, customer records, private account pages, screenshots, full names, private handles, or payment details. The generated brief should use placeholders only.