Skip to main content
FinalRun ships three AI coding agent skills that let your AI coding agent generate tests, run them, and fix failures — all from chat. The skills work with any agent that supports the skills protocol, including Claude Code, Cursor, Windsurf, and similar tools.

​
Install the skills

Run the following command once in your repository to install all three skills into your coding agent:
These skills work with any AI coding agent that supports the skills protocol. You do not need to configure them separately for each tool.

​
Available skills

/finalrun-generate-test

Reads your app’s source code, infers the app identity, and generates complete YAML test specs with setup, steps, and expected state — organized by feature folder.

/finalrun-use-cli

Validates and runs tests once they are generated. Handles flag selection, env binding, and post-run artifact inspection.

/finalrun-test-and-fix

Orchestrates the full generate → run → diagnose → fix loop. Keeps iterating until the run is green or a genuine blocker is hit.

​
/finalrun-generate-test

The /finalrun-generate-test skill reads your app’s source code, infers the app identity (package name or bundle ID), and generates complete test specs organized by feature folder under .finalrun/tests/. Example usage
When you invoke this skill, the agent works through the following sequence:
1

Read source code

Reads relevant application source code to understand the UI, user flows, and infer the app’s package name or bundle ID.
2

Set up configuration

Creates or updates .finalrun/config.yaml with the app identity and scaffolds environment bindings under .finalrun/env/.
3

Propose a test plan

Presents the proposed test files, feature folder structure, and environment binding strategy for your approval before writing any files.
4

Generate YAML specs

After you approve, writes YAML specs under .finalrun/tests/<feature>/ and a matching suite manifest under .finalrun/suites/.
5

Validate with finalrun check

Runs finalrun check to validate the workspace, bindings, and generated specs. Fixes any issues until the check passes.
The agent will not write test files until you explicitly approve the proposed plan in step 3. Review the proposed paths and binding strategy before confirming.

​
/finalrun-use-cli

The /finalrun-use-cli skill validates and runs your tests once they are generated. It handles CLI flag selection, env binding, and post-run artifact inspection — including reading result.json, screenshots, and device logs on failure. Example usage
The agent validates the workspace with finalrun check before executing, explains the exact command it will run, and summarizes the outcome with artifact links when done.

​
/finalrun-test-and-fix

The /finalrun-test-and-fix skill orchestrates the full generate → run → diagnose → fix loop. It calls /finalrun-generate-test to author tests, runs them via /finalrun-use-cli, reads the CLI artifacts on failure, classifies whether the bug is in the app code or the test spec, applies the narrowest possible fix, and re-runs until the run is green. Example usage
The agent stops the loop only when the run is green, or when execution is genuinely blocked — for example, no emulator is available, a required secret is missing, or you explicitly opt out. In a blocked state it prints the exact command for you to run locally.
Use /finalrun-test-and-fix as your default entry point when you finish a UI feature. It covers the full cycle without needing to invoke the other two skills manually.

​
Auto-triggering FinalRun after feature work

You can configure your AI agent to automatically generate and run FinalRun tests whenever it finishes a UI feature — without you having to ask explicitly. Add the autotrigger content to your AGENTS.md file:
1

Install the skills

2

Add the autotrigger rule to AGENTS.md

Add a definition-of-done rule to your AGENTS.md that instructs the agent to run the /finalrun-test-and-fix skill automatically after every UI change. The rule tells the agent not to mark a task as done until FinalRun coverage is updated and the run is green.
With the autotrigger in place, your agent treats a passing FinalRun run as a hard requirement for every UI task — no separate prompt needed.