Skip to content

Keep your app on track

You build the app. Skeg runs it on a real device after each change and tells you whether it still does what it should. This section shows how to wire that into your daily work.

The loop

graph LR
    A[Change code] --> B[Build and install]
    B --> C{Behavior meant<br/>to change?}
    C -- No --> D[Replay stored plans]
    C -- Yes --> E[Goal run with<br/>acceptance criteria]
    E -- passed --> F[Promote to plan]
    F --> G[Plan joins the<br/>regression suite]
    D -- fail --> A
    D -- pass --> H[Merge]
    G --> H
    H --> I[CI replays the suite<br/>by tag]
You changed… Skeg step Needs a model?
Code, and behavior should stay the same Replay the stored plans for the screens you touched. No
Behavior on purpose (new screen, label or flow) Goal run against acceptance criteria, then promote the passing run to a plan. Yes
Behavior on purpose, short flow Record by hand (tap, type, assert), then save it as a plan. No
Nothing yet. You want to see the screen Screenshot through a live inspector. No

Stored plans are your regression baseline. Every promoted plan is one more flow that CI and your coding agent can replay with no model and no cost per run.

Where to start

  1. Make your app testable: stable element ids, a known start state, test data.
  2. Start with your coding agent: a ready kit (MCP config, skill, first prompt). Coding agents (MCP) explains the tools behind it.
  3. Run Skeg in CI: replay a tagged suite on every pull request.
  4. Maintain your plans: read failures, re-record intended changes, port to a second platform.