liveBook · Manning

Whitelist approval

As explained in Build an AI Agent (From Scratch)

Whitelist approval is an approval policy in which only explicitly approved pending operations are executed and all others are rejected.

Whitelist approval executes only explicitly approved pending tool calls. The agent pauses before execution, stores approval-required calls in execution state, and rejects unapproved calls automatically. section 6.3.7

Partial approval

When an agent requests three tool calls, a user can approve two and reject one; only the two approved calls execute. This prevents irreversible actions such as file deletion or payment processing from running before consent. section 6.3.7

How Build an AI Agent (From Scratch) covers it

How deep it goes
Explains deferred execution, state storage, confirmation matching, and result creation, without detailing dependent-call resolution.
Examples
Uses three requested calls, selective approval, file deletion, and payment processing to illustrate the policy.
What it assumes
Assumes tool calls, confirmation metadata, execution context, and agent state; distinguishes approval-required tools from immediate tools.
What you can do after
Apply it via an agent confirmation flow that pauses, resumes, and executes only approved pending calls.
How it connects
Connects pre-execution blocking, tool calls, pause and resume, callbacks, and ToolResult generation.
Cover of Build an AI Agent (From Scratch)
liveBookmanning.com
Read Build an AI Agent (From Scratch) on liveBook Get the book at manning.com

This explainer was generated by Manning's knowledge graph from the text of Build an AI Agent (From Scratch). The section links point to the parts of the book it draws on.