START HERE
What you will learn
- Follow a website change from source code to a public response.
- Understand the difference between a build, an upload and a verified release.
- Identify where permissions and recovery decisions belong in an agent-assisted process.
Before you start: You should know that a repository stores source code. The other terms are explained below.
What is the agent actually doing?
Here, agent-assisted means software helps edit source, run commands, inspect results and coordinate a release under the user’s instructions. It does not mean the agent should have unlimited authority or that a successful command proves the website is correct.
The website runs on a remote virtual machine, or VM: a server provided as a software-defined computing instance. The build turns source code into files the server can serve. The deployment then transfers and activates those files. These are different stages, and each can succeed or fail independently.
What is verified in the current MintBios setup?
The implemented release path uses GitHub Actions for validation and deployment, a dedicated SSH deployment identity, commit-specific release directories and public HTTPS checks. SSH is the encrypted connection used for the upload and release command. The release identifier lets us check which source revision the server is serving.
VPN-only management and Terraform-controlled infrastructure are separate proposed patterns described in the related guides. They are not prerequisites already implemented by this release script. During service or configuration problems, a verified SSH deployment has also been used and recorded as a manual release path.
Step-by-step process
Record the change in source control
A commit identifies a particular source revision. Keeping the change in the repository makes it possible to review it and connect the deployed output to its origin.
Run validation and build the files
The workflow runs tests and creates the static website files. Link and metadata checks inspect the generated pages. These checks do not replace a visual browser review.
Package the release
The files are placed in an archive associated with the commit. Preparing one release package avoids editing individual live files in place.
Verify the SSH destination and upload
The client requires the expected server key and authenticates using the deployment identity. A host-verification failure is a reason to investigate, not to disable the identity check.
Activate the versioned directory
The release script extracts a commit-specific directory and switches the server’s current-release reference to it. A local content check can restore the previous reference if activation fails.
Check the public website
Check the release identifier and expected content over HTTPS. Public DNS, certificates or page behaviour may fail even when the local content check passed. A later public failure is reported; it does not automatically trigger the local script’s rollback.
FOLLOW ONE EXAMPLE
Worked scenario
Publishing a change to a research article
This scenario follows the implemented release design. It explains the stages rather than claiming every release always passes them.
- The source change updates the article and receives a commit identifier. Validation checks the generated pages, then the release package is uploaded.
- Activation succeeds and the local check returns the expected identifier. The public article URL is then opened through HTTPS and checked for its new text.
- If the public URL still shows the old text, inspect the identifier and the request path before declaring success. If a recovery is necessary, deliberately select the previous known-good release and verify it.
What this shows: A completed deployment means the intended revision is available at the public URL and relevant checks have passed. “The build passed” describes only an earlier stage.
The process at a glance
Read Mermaid diagram source
flowchart TD
A[Source commit] --> B[Tests and build]
B --> C[Release archive]
C --> D[Verified SSH upload]
D --> E[Activate versioned directory]
E --> F{Local content check passes?}
F -->|No| G[Restore previous release]
F -->|Yes| H[Check public HTTPS and content]
H --> I[Record result or investigate]What you should have at the end
- A traceable connection between source revision, release package and public content.
- Separate evidence for validation, upload, activation and public verification.
- A recovery decision based on the stage that failed.
Common mistakes and how to avoid them
Calling a successful build a successful deployment
The files may never have reached or been activated on the server.
Giving release automation full administration access
Use the permissions needed for the release; keep host administration separate.
Assuming every failure rolls back automatically
Know which check triggers rollback and which failures need a separate decision.
APPLY THE IDEA
Try it yourself
Your task: Sketch the path for one proposed website change. At each stage, write the evidence you would need to call it successful.
What to produce: A release checklist covering source revision, build, destination, activation and public content.
Check your understanding: If the public page were wrong, could you identify the last successful stage without exposing credentials?
Key terms in plain English
- VM
- A virtual machine: a server instance running remotely.
- Build
- The process that turns source code into deployable files.
- Release
- A particular version prepared for use on the server.
- Rollback
- Returning to a previous known-good version.