Security
deployhealth looks at your code's configuration and at your clients' servers, so here is exactly what it stores, what it does, and how to reach us. The code is public, so you can check every claim below.
What we store
- Variable names and file:line, never values. For each deploy: the commit sha, branch and time, and for each finding the variable name, the file and line that reads it, and the env file involved, plus the names each env file defines. The scanner runs in your CI and reads env files only for their names; the ingest API accepts only names, paths and line numbers.
- Findings per deploy (missing, unused, out of sync), so you can see what changed.
- Endpoint URLs you add, with their names and settings, and for each check its time, status code, latency and a short error reason. Raw checks are deleted after 30 days; daily totals are kept for monthly reports.
- Alerts (when they opened and resolved, and the message) and your alert webhook URL. Webhook URLs contain secrets, so logs show only their host.
- Ingest tokens only as a SHA-256 hash. The token itself is shown once, when it's created in the project's settings.
- Your account: your GitHub id, login, name, email address and avatar URL, from GitHub sign-in. Sign-in reads your GitHub profile (id, login, name and avatar) and your email address, even if it is private on GitHub (GitHub scopes read:user, user:email). Sign-in gets no access to your repositories (the optional GitHub App below is separate). Clients, deploy notes and contact emails are what you type in.
Error reports and logs
When the servers hit an error, they send a report to Sentry: the error's type, its message and stack trace (files and lines in deployhealth's code), the release, and the route's pattern (such as /projects/[projectId], never the address you visited). Before it leaves the server, request bodies, headers, cookies, query strings, who was signed in, local variables and source lines are removed, and tokens, webhook URLs, email addresses, long random strings and anything shaped like NAME=value are replaced with [redacted]. Server logs are redacted the same way.
Backups
Once a week the whole database is copied, compressed, to a private Cloudflare R2 bucket used only for backups, where it is encrypted at rest. The last 4 weekly copies are kept and older ones deleted, so data you delete leaves every backup within about 4 weeks. Next to them, a list of deleted accounts (GitHub numeric ids and dates, the last 35 days) is updated daily, so restoring a backup re-applies every deletion made since it was taken.
Exporting and deleting your data
On your Account page, "Export my data" downloads a JSON file of everything deployhealth holds for you (no tokens, no webhook URLs), and "Delete my account" deletes your account and everything it owns at once, after you type your GitHub login. Deleting removes the same rows, in one database transaction, as a deletion we run on request: your account, clients, projects, deploys and scans, endpoints and their checks, alerts, and the GitHub App installations linked to you or installed from your account, with their pull request checks.
A deletion leaves one record: your GitHub numeric id and the date, kept for 35 days so that restoring a backup can never bring the account back.
How checks run
- One request per check (GET or HEAD) with a 10-second budget, following at most 5 redirects.
- SSRF-guarded. A URL must be public http(s): no credentials, no private, loopback, link-local or otherwise reserved addresses, nor IPv6 addresses that carry or tunnel to an IPv4 one. That's checked when you save it, and again at connect time for every redirect hop, which also defeats DNS rebinding. Alert webhooks go through the same guard.
- Response bodies are never read or stored: only the status code and timing.
- Checks are spaced per hostname: one hostname is checked at most once every 10 seconds, however many accounts monitor it. Different names for one server (another subdomain or alias, or its IP address) are spaced separately. An account can monitor at most 500 endpoints (100 per project), with at most 50 checks started per minute. The spacing applies to the hostname in the saved URL: requests that follow a redirect (up to 5 per check) aren't counted toward the spacing of the host they land on.
- Checks of one hostname start at least 10 seconds apart, so more than 5 endpoints on the same hostname are checked less often than every minute.
The GitHub App (pull request checks)
If you install it, the App asks GitHub for read access to the contents of the repositories you choose, and write access to their pull requests and checks (to comment and add the deployhealth / env check). For each pull request it reads, at the pull request's head and at its merge base, only the files the scanner reads (source files in the scanned languages up to 512 KB, env files, Docker Compose files, Spring's application and bootstrap config, Symfony config YAML and Laravel's artisan file, whose contents aren't read; never tests, fixtures, vendored code or build output such as Laravel's storage/ and bootstrap/cache/ or Maven's and Gradle's target/ and build/), plus the pull request's diff and commit messages. From config files it keeps only variable and key names, never values. At most 2,000 files and 20 MB per pull request, fetched only from api.github.com.
The scanning runs apart from uptime checks, in a separate thread, and stops after 60 seconds of work: a pull request that takes longer, or whose files can't be scanned, gets a neutral “Couldn't be checked” check run and no comment, and isn't retried. A slow pull request can't hold up anyone's uptime checks or alerts.
Strict mode's limit: a pull request that couldn't be checked (the check crashed, ran out of time, or hit an error on our side) or was too large to check (past the limits above, or a repository GitHub can't list in one tree) gets a neutral check run, even in strict mode, and GitHub counts a neutral check as passing. Strict mode blocks a merge only for what a check actually found.
It stores what it reports: variable names with file:line, the names of committed env files, how many secret-shaped strings it saw (never the strings), the pull request number, its author and whether a coding agent wrote it. File contents are discarded as soon as the check finishes. Webhook deliveries are verified with a shared secret before anything is read, and uninstalling the App deletes its pull request checks.
How share links work
A monthly report's share link isn't stored anywhere: the link itself carries the client, the month and an expiry date 90 days out, signed with HMAC-SHA256 using a key only the server has. Anyone with the link can read that one report until it expires, and nothing else; changing any part of it breaks the signature. Links can't be revoked one by one: rotating the key revokes all of them at once. Shared pages are rate-limited per IP address and ask search engines not to index them.
Report a vulnerability
Please report security issues to security@deployhealth.dev, not in a public issue. Include what you found and how to reproduce it. Testing against your own account and your own endpoints is welcome; please don't access other people's data or degrade the service.
Machine-readable: /.well-known/security.txt.
Our commitment
If a security incident affects your data, we will email every affected user within 72 hours of confirming it, at the email address on their GitHub account, saying what happened, which data was involved, and what we are doing about it.