Fit RunStory into your workflow
RunStory checks your software by running it. Start that check where it suits the way you work: at each commit, over a span of commits, at each push, or in your pipeline.
Where a check can start
Four places to start a check. Three go through the runstory CLI: two on your machine, and one in a pipeline job with an API key. The GitHub App needs no CLI at all.
| A check starts | Driven by | How |
|---|---|---|
| At each commit | The CLI, on your machine | Turn it on with runstory settings --automatic on |
| Over a span of commits | The CLI, on your machine | You or your coding agent runs runstory run <from>.. |
| At each push | The GitHub App | Connect the repository and turn on automatic checks. Each push is then checked after CI passes |
| As a pipeline step | The CLI, in a job | The job runs runstory run, signed in with an API key |
Wherever it starts, it's the same check. RunStory builds and starts your app from the commit, uses it the way a user would with the latest frontier models, and confirms each failure before it reports it.
Find your workflow
Check every commit, or a whole span of commits when you ask.
The GitHub App checks each push and reports on the pull request.
One check in your pipeline gates the merge into main.
Check each agent's branch before it merges back.
The loop's script starts each check and hands failures to the next turn.
Check everything since the last release before you tag the next.
Why every workflow needs it
Coding agents write code faster than anyone can run it. The checks most teams already have read the code or run its tests.
| The check | What it tells you |
|---|---|
| Tests in CI | The cases that have a test still pass |
| Code review, by a person or a bot | The diff reads well |
| The coding agent's own report | The agent believes the work is done |
| RunStory | What the app does when someone uses it |
Spotify's engineers call one failure of their background coding agents the most serious: a pull request that "passes CI but is functionally incorrect."
However bespoke your software factory is, it needs that last signal somewhere on the line, from something other than the agent that wrote the change. Where it goes depends on how you work.
What comes back
Every workflow closes the same loop. A confirmed failure goes back to your coding agent with a script that repeats it, the agent commits a fix, and RunStory runs the script again.
| Where | What you get |
|---|---|
| Your coding agent | Each new finding at the end of its turn, in Claude Code, Codex, and Cursor |
| Your terminal | runstory status and runstory findings, or JSON for a script |
| The check's web page | Each finding and its script, plus a video of the browser for a web app |
| GitHub | A Check named Runstory on the commit and its pull request |
| Your pipeline | An exit code: 0 for no confirmed problems, 1 for a new one |
Get started
RunStory is in beta for CLIs and web apps, and your first 10 checks are free.
Set up with your agent
Set up runstory for me. Fetch https://runstory.specstory.com/
Or install it yourself:
-
Install the CLI on macOS or Linux.
curl -fsSL https://runstory.specstory.com/install.sh | sh -
Set up a repository and run your first check. Each command runs only if the one before it worked, so nothing is checked until your app boots.
runstory init && runstory setup && runstory run
For a pipeline, or any machine that runs without you, create an API key on the keys page and store it as the secret RUNSTORY_API_TOKEN.