liveBook · Manning

Progressive disclosure

As explained in Designing AI Agents

Progressive disclosure is a technique that reveals detailed information or capabilities only when they become relevant to the current task.

Progressive disclosure reveals information or capabilities when they become relevant to the task. In an agent runtime, it limits the context and selection noise created by large catalogs: expose a compact name and description first, load a full contract for a promising candidate, and retrieve detailed references only when needed. section 6.3.2

Agent Skills

At session start, Claude Code exposes visible Skill names and, subject to listing budgets and visibility settings, descriptions. When a task uses a Skill, its full SKILL.md loads; referenced files, such as database schemas or API documentation, load on demand. The YAML header, markdown body, and referenced files form discovery, instruction, and detail tiers. section 3.1.3 section 3.4.5

The savings depend on the actual descriptions and listing settings, not just the size of the deferred content. Use the context inspector to measure the current configuration. Claude API tool search applies a similar approach: deferred tool definitions remain outside the initial model context and expand after a search result. section 3.1.3 section 6.3.2

Disclosure is not authority. Loading a deployment tool does not grant permission to call it; eligibility and execution controls remain separate. For an action, prefer a typed API or admitted Skill when it expresses the operation directly, and verify the resulting state when using less structured interfaces. section 6.3.2

How Designing AI Agents covers it

How deep it goes
Explains staged loading from descriptions through full contracts to references, while distinguishing disclosure from execution authority.
Examples
Illustrates disclosure with Agent Skills and deferred Claude API tool definitions.
What you can do after
Evaluate progressive disclosure by measuring listing costs and separating capability loading from permission to execute.
How it connects
Connects progressive disclosure to context windows, tool search, and authorization.

Also explained in

Compare how 2 books explain progressive disclosure →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Designing AI Agents. The section links point to the parts of the book it draws on.