
Bundling hundreds of skills into every agent bloats its context. An agent skill avoids that by packaging a set of instructions, with optional scripts and documentation, that an agent loads only when a task calls for it. Every skill includes a SKILL.md file containing a name and a description in its frontmatter, and the agent uses that description to decide whether the skill is relevant. For the Professional Agentic Architect exam, understanding how skills are packaged and retrieved through Skill Registry helps distinguish on-demand instruction management from static bundling and conversational memory.
Skill Registry is a secure, private, low-latency, central repository for agent skills that agents can search at run time. Skill Registry is in Preview, and it provides a central way to validate, version, and retrieve skills without tying skill updates to individual agent builds:
SKILL.md file with a name and description in its frontmatter, and Skill Registry validates every package.RetrieveSkills method uses each skill's display name and description to find the skills relevant to a request, allowing agents to load only what a request needs.gcp-skill-registry skill that lets agents search for and manage skills.Because the agent uses the description in SKILL.md to decide whether a skill is relevant and RetrieveSkills matches requests using each skill's display name and description, clear skill metadata determines which skills get loaded for a given request. Storing each revision as an immutable snapshot with a default revision pointer keeps version history intact in one central place.
On the Professional Agentic Architect exam, Skill Registry is often contrasted with simpler storage or deployment patterns that miss key requirements. Bundling every skill into an agent or copying every skill into each image repeats the context bloat problem and ties every skill update to every agent's build.
When skills are loaded matters as much as where they are stored. Loading every skill when an agent starts brings the context bloat back, since the agent loads skills whether or not a task calls for them. Plain file storage can hold skill packages, but it offers no search by intent and no revision history, and those are exactly what Skill Registry adds.
Memory Bank serves a different role altogether. Memory Bank stores memories generated from conversations, not versioned packages of instructions, code, and documentation. Skill Registry manages reusable instructions that agents retrieve by relevance, while Memory Bank keeps facts drawn from past conversations.
Our Professional Agentic Architect practice exams test agent skills and Skill Registry alongside Memory Bank and context management, with detailed explanations that drill these distinctions.