

Each app gets its own release workflow per slot, filtered to its source path and started by a push to that slot’s branch. The run builds the image once, pushes it to the registry and brings the service up on the target machine, one release at a time. The same pattern covers a cloud VM, rented hardware or a machine you own.

Blue and Green Slots
Each app runs two slots from the first deploy, green for your customers and blue for the next version under test. Both share one machine, so the candidate costs nothing extra. Promotion moves the exact image tested on blue to green, so rollback is a single step. A release counts once the right image runs and its address answers.

Claude Code as Harness
Claude Code is the harness: it writes the workflow, the service definition and the database change, runs the checks and reports the release it verified. Guard hooks and an approval gate set what runs unattended. Agents ship through the same pipeline as apps, each with its own identity. GitHub Copilot coding agent works in the same repository.
Stays Live
customers keep using the running release while the next version is tested
Identical Build
the version customers get is the exact candidate build that already proved itself in testing
Faster Releases
every push to the release branch starts the pipeline automatically, keeping new features on a steady cadence
Candidate Included
the tested candidate release runs on the same machine, inside the hosting you already pay for
One Pipeline for Apps and Agents
Build and Ship 4 steps
01
One Workflow Per App and Slot
Each app gets its own GitHub Actions workflow per slot, filtered to that app's source path and triggered by a push to the branch the slot builds from, so a push is all it takes to release. Claude Code writes and maintains these workflows in the repository, so the pipeline is reviewed like any other code.
02
Build Once, Run the Same Image Everywhere
The workflow builds the image once, pushes it to the registry, then pulls and starts that exact image on the target machine. The tag on the box always matches the tag just pushed, per service, so what runs is what was tested.
03
Proof of Every Deploy
Every run checks three facts before it counts as done: the image on the box carries the digest just pushed, the running container matches that image, and the public hostname answers.
04
One Deploy at a Time
Every run builds in parallel and then queues on a lock held on the target machine before it touches anything there. Concurrent releases wait their turn. The lock on the box protects every deploy that is under way.
Blue and Green by Default 3 steps
05
Blue and Green From the First Deploy
Each app runs two slots from day one, each with its own hostnames and environment file: green for customers from the main branch and blue for the candidate. Both share one machine, so the candidate costs nothing extra.
06
Promotion by Retag
Images carry no environment-specific detail, so promotion retags the exact image that ran on blue. An immutable tag then makes rolling back a single step.
07
Secret Changes
Changing a secret or a connection string goes through the secrets MCP server, with a human in the loop. The container is then recreated so it picks up the new value. A new hostname is added once its DNS record resolves, so certificate validation runs straight through.
Agents in the Pipeline 2 steps
08
Agents Ship Through the Same Pipeline as Apps
An MCP server, an agent service or a scheduled worker counts as one more app in the estate: it gets its own workflow, its own two slots and the same release checks as everything else. A tool an agent calls proves itself on the candidate slot before it reaches the version people rely on.
09
Claude Code in the Pipeline
Claude Code works here as the harness: it writes the workflow and the compose service, runs the checks and reports the digest it verified, with guard hooks and an approval gate deciding what may run unattended. GitHub Copilot coding agent fits into the same repository for teams that use it. Matching this to your own pipeline starts with a look at what you deploy today.
Frequently Asked Questions
What does proving a release involve?
Three checks run in sequence. The digest of the image just pushed is compared against the image on the box, which establishes that the right build is there. The running container's image id is compared against that image, which establishes that the running process is the one deployed. The public hostname is called and must answer.
Does blue/green mean paying for a second environment?
Both slots run on the same machine, with their own containers, hostnames and env files, so the candidate environment adds no separate hosting line. Databases sit on one shared instance. The candidate slot exists from the first deploy, which keeps promotion a retag of the tested image.
Do agents get to deploy unattended?
Per environment, and the boundary is written down before the first run. Read-only checks, builds and candidate-slot deploys run unattended. Anything touching the production slot passes an approval gate. Guard hooks enforce the limits. Every action carries the agent's own identity, so the audit trail names the agent that acted.
How do secrets and connection strings get changed?
Secrets and connection strings are managed through a secrets MCP server, with a human in the loop before a value changes. The container is recreated afterwards, because the env file binds when the container is created. Values stay out of the image and out of every procedure file, which names the file that holds a value.
