Operate
Project doctor
One ranked list of everything in a project that needs attention, from every part of CLI, each with a fix you can apply in one click.
The project doctor looks over everything in a project (deploys, health checks, variables, domains, backups, credit, costs, leaked secrets) and gives you one list of problems, worst first, at the top of the project page. Each says what's wrong in plain words, often with a button that fixes it.
How to use it: open a project and read the list at the top. When it's empty, nothing needs you.
CLI has a lot of moving parts: deploys, health checks, variables, domains, backups, credit, browser checks, workflows, leaked secrets, costs. The project doctor looks at all of them and gives you one list, worst first, at the top of the project page. Each item says what's wrong in plain words and how to fix it, often with a button that does it.
What it checks
| Area | Raises |
|---|---|
| Deploys | The last deploy failed (critical if nothing is serving); a static site that was never published |
| Health | An app is down; an app was down for a while today; an app with no health check path |
| Variables | A service refers to a variable that doesn't exist, so the app would get an empty value |
| Security | Leaked secrets that haven't been rotated (/docs/secret-scanning) |
| Domains | A custom domain whose DNS records still aren't in place |
| Data | A database with no backup in two days, or never |
| Money | Credit running out within a week; the month forecast over budget; a spending spike |
| Monitoring | A browser check failing twice in a row |
| Workflows | Workflow runs that failed today |
| Cost | Changes that would save at least a dollar a month; apps running at their plan's limits |
Each item is Fix now (critical), Soon (warning) or a Tip. The health score starts at 100 and drops by 25 for each critical item, 8 for each warning and 2 for each tip.
Fixes
Where a fix is a single safe action, the item has a button: back up a database now, check a domain again, retry a failed workflow run, apply a right-sizing change, run a browser check, deploy an unpublished site. Everything else links to the page where you fix it.
GET /v1/projects/:id/doctor{
"score": 59,
"summary": "1 thing needs attention now, and 2 more soon.",
"issues": [
{
"id": "backup:svc_db",
"severity": "critical",
"area": "data",
"title": "postgres has never been backed up",
"detail": "If the database is lost or a migration goes wrong, this is what you'd restore from.",
"serviceId": "svc_db",
"fix": { "label": "Back up now", "request": { "method": "POST", "path": "/v1/services/svc_db/backups", "body": {} } }
}
]
}An item's id stays the same while the problem lasts, so it can be tracked or dismissed. fix.request is the exact API call that fixes it; fix.dashboardPath is the page to go to otherwise.
For agents
The skill tells agents to check the doctor (cli_doctor) when someone asks how their project is doing and after any change. They work through critical items first, and ask before a fix that costs money or touches a live app.