Where Learn more appears, and why
Some screens in Context carry a small Learn more link. It opens the Resources page that answers that screen’s question, at the exact heading. We added these links where people make a decision they can’t undo by looking, or write something an agent will act on. Everywhere else, the screen stays clean.
The rule
A screen gets a link only when at least one of these is true:
- The choice has effects you can’t see on the screen. Who can see a file, what an agent may do without asking, whether to set up a benchmark or a monitor.
- You’re writing something an agent will act on. An agent’s instructions, a skill, an automation that runs while nobody watches, a dataset.
- The screen expects a format you couldn’t guess. The columns of a dataset file, or how to read a run’s scores.
Every link works the same way:
- It opens the answer, not a page top. Each link points at a heading.
- It stays out of the way. One link per screen, small, next to the title or main button.
- It never loses your work. In a dialog or an unsaved form, it opens in a new tab.
Where the links are, by section
Drive and sharing
Where a file lives and who it’s shared with decide which people, and which agents, can see it.
| Screen | Opens |
|---|---|
| Share dialog | Share work safely › Roles |
| Drive, beside New | Find your work in Drive › Add and organize files |
Agents
Most of an agent’s quality comes from what you write when you create it, and its permissions decide what it can change without asking.
| Screen | Opens |
|---|---|
| Agents, beside New agent | Meet your agents |
| Create agent | Build an agent › Create the agent |
| An agent’s Permissions | Decide what agents can do › Presets |
| Review approvals | Watch, steer and finish › Answer approvals |
Automations and skills
Both repeat whatever you write, on every run or every task, so a mistake in the instructions repeats too.
| Screen | Opens |
|---|---|
| Automations, beside New | Recurring reports with automations |
| Create automation | Recurring reports › Schedule a report |
| Skills, beside New Skill | Turn repeat work into skills |
| Skill editor | Turn repeat work into skills › Create a skill |
Applets and workspaces
An applet’s brief is all the agent has to build from, and a new workspace sets who sees what by default.
| Screen | Opens |
|---|---|
| New applet | Build an applet › Build one |
| New workspace | Organizations and workspaces › Join or create a workspace |
Evals
Evals asks for several linked choices before anything runs, and its files and scores have formats you need to know.
| Screen | Opens |
|---|---|
| Evaluations | Benchmark or monitor |
| New evaluation | Benchmark or monitor › Create a benchmark |
| Datasets | Dataset file templates |
| Import tasks | Dataset file templates › File types |
| Shared rubric | Writing criteria that grade well |
| Run scorecard | Reading results › A run |
Where there’s no link, on purpose
Home, Inbox, the message box and the settings lists have no link. Nothing is decided there, and a link on every screen would teach people to ignore the ones that matter. Every page is still one click away: Resources has its own item in the left rail, and ⌘K finds any page by name.
Keeping the links honest
All 18 links live in one map in the docs repo (docs/links.json). The docs checks refuse a link to a page or heading that doesn’t exist, and the product’s checks refuse a link from a screen that doesn’t exist. When a redesign makes a screen explain itself, its link comes out.