Plans and runs¶
Concepts¶
| Term | Meaning |
|---|---|
| Project | The container for everything below. Members see the same data. |
| Source | A ticket or requirement: title, description and acceptance criteria. Plans link to sources. |
| Plan | An ordered list of steps (launch, tap, input, clear text, back, assert, swipe, server) for one platform. |
| Environment | Named variables. A step uses one as ${name}. Mark a variable secret to hide its value. |
| Device target | Platform, device and app id. Saved in Setup. A plan can name a default target. |
| Run | One execution of a plan, or one goal-agent attempt, with its steps, evidence and verdict. |
| Captures / Library | Captured screens and reusable steps for building plans. |
Writing plans¶
- Record taps in the Inspector (Chat → Inspector → Record → Save as plan).
- Edit steps in Plans. Each step has an action, a selector and an optional value.
Do not use wait steps
The step editor offers a wait action, but replay does not support it: the step fails as
"unsupported action". Give the next step a longer timeout instead.
- Generate a plan from a goal in Chat. See Goal agent.
Selectors take one of these forms: id=<resource id>, text=<visible text> or
point=<x>,<y>. Prefer id=: it survives layout and copy changes.
A step that names an element (id= or text=) replays against that element only. If the
element is gone, the step fails. It never falls back to tapping the recorded point, because a
blind tap would pass when the element is missing. Only a step recorded as a point= taps a
screen position.
Tags and suites¶
Give plans tags in the plan details sheet (comma separated, for example smoke, checkout).
In the Run dialog, pick a tag to run every enabled plan with that tag in order. All plans in
one suite must share a platform and a target.
Environments¶
- Open Envs, create an environment, and add variables.
- Reference a variable in a step value as
${username}. - Choose which environment a plan uses by default in the plan details sheet. The topbar switcher picks the environment for plans with no default.
Warning
Do not store production credentials in an environment, even as a secret.
Running¶
- Run plan in Plans executes one plan. The Run dialog lets you pick a tag instead.
- When an Inspector session is live on the same device, the run replays through that session. The Inspector shows a "plan in progress" overlay until the run ends.
- Stop cancels a run in progress.
Run history¶
Runs lists every run, newest first. Filter by status, text and origin:
| Origin | Started by |
|---|---|
| Human | You, from the app |
| Agent | Skeg's goal agent |
| MCP | An external AI agent through a project token |
| CI | The runner CLI in a pipeline |
The default view shows human runs. Open a run to see each step, its screenshots and its actions.