Talents
Choose the expert instructions your AI agent should use.
A talent gives the AI agent instructions for a specific capability. For example: organizing a change or checking a test.
Where do talents come from?
| Source | Purpose | Who manages them? |
|---|---|---|
| Shared catalog | Reuse available expertise. | Catalog maintainers. |
| Your organization | Add instructions specific to your team. | Your organization's owners and administrators. |
Find expertise
Under Talents, search for a name, identifier, or technology. The most relevant
results appear first; matching tags explain the suggested associations. For example,
GitHub Actions and github-actions find the same talents. Accents and capitalization
do not affect the results.
You can complete the last word or make a small typo in a long word:
kubernets finds Kubernetes. All entered words must match; an unknown
search can return no results. Filters remain available.
The talent page first shows a quick summary, then a public description explaining the context and expected benefit. It also shows the version, source, and amount of context its three versions can pass to the AI agent: short, detailed, and complete. The short version is currently used. Each amount counts tokens in the corresponding instructions; this lets you compare their size without revealing their contents. An organization talent only shows the size of its published instruction. Required talents appear as added automatically, whether they are prerequisites or internal parts of a composition. Suggested additions remain optional. Execution instructions, examples, and tests are reserved for AI agents working on active projects: they are not shown in the dashboard.
The Installation section lists projects that already have the talent and offers only other projects as destinations. You can select one or more projects before confirming. Each project is handled separately: if one installation fails, previous successful installations remain in place and Grace lets you retry only the failures. The page also lists specializations that contain this talent, so you can choose between installing one capability and installing a coherent set.
Available, installed, or active?
A talent in the catalog is available. An installed talent belongs to the project's selection. Only active talents can be served to the AI agent. The AI agent designates the expertise its task needs; Grace can also include its dependencies and your organization's mandatory instructions.
Why dependencies?
An architecture talent may rely on several more specific rules. Grace identifies missing prerequisites so your team can complete the selection. A recommendation is a suggestion, not a prerequisite.
For a specific capability, install a talent. For a set of capabilities, use a specialization. For your team's own instructions, see organization talents.