Progressive Disclosure
Reveal detail only when it becomes relevant.
A context strategy where the agent first sees lightweight summaries or pointers, and loads the full content only when needed. Keeps the window small while keeping depth available.
The alternative — load everything up front — fills the window with material that mostly does not matter and dilutes attention on what does. Progressive disclosure is how you get both breadth and focus.
This is the answer to "why not just put everything in the context". Because attention is a budget that every token competes for, a catalogue of summaries with the full detail one call away beats a wall of documentation that is technically all present and practically half-read.
A catalogue of tools instead of all of them
A platform exposes two hundred API operations to its agent.
Hand it a searchable index of operation names rather than two hundred full schemas. It loads the two it needs, and the action-selection problem gets simpler rather than harder as the surface grows.
- blogEquipping agents for the real world with Agent Skills ↗
Defines Skills as discoverable folders of instructions, scripts and resources loaded progressively — the modular-reusable-procedure half of agent scaffolding.
- blogEffective context engineering for AI agents ↗
Defines context engineering as curating everything supplied at inference, names context rot, and covers just-in-time retrieval, compaction and keeping the window high-signal.
- docsTool search ↗
Deferred tool loading and namespaces: the model searches for tool definitions instead of being handed every schema up front.
Context Pointer · Reference
A cheap reference to something the agent can fetch if needed.
Skill · Capability
A packaged procedure the agent can load when the task calls for it.
Context Engineering
Deciding what goes in the window, in what order, at all.
AGENTS.md · Project instructions file
A plain file in the repo that tells agents how this project works.