ci: add Gitea Actions workflow ported from .gitlab-ci.yml

Merge branch 'ci/release-permalink' into 'main'
Give releases a stable latest-download URL See merge request coffey-labs/ihasmail-oneshot!2
2026-09-21 22:44:22 -07:00 · 2026-09-20 21:15:14 -07:00 · 2026-09-20 20:58:45 -07:00 · 2026-09-20 20:35:35 -07:00 · 2026-09-20 20:13:35 -07:00 · 2026-09-15 12:31:19 -07:00
17 changed files with 1216 additions and 837 deletions
@@ -0,0 +1,103 @@
 # CI on the self-hosted Gitea, ported from .gitlab-ci.yml during the move off
 # GitLab (2026-09-22). Gitea reads .gitea/workflows and ignores .github/ once
 # this directory exists; .github/workflows stays as it was for GitHub.
 #
 # Every job runs in an image pinned by digest (tag in the trailing comment),
 # and the only action used is coffey-labs/actions/checkout pinned by SHA. The
 # instance resolves short `uses:` against itself, never GitHub, so nothing
 # unreviewed can be pulled in.
 #
 # The shape is the same as before -- tag-driven, amd64 and arm64,
 # reproducible. GitLab needed a generic package registry plus release-cli
 # links; Gitea attaches the tarballs to the Release itself, as GitHub did, so
 # the build and the release are one job and nothing is handed between jobs.
 #
 # e2e.yml is deliberately NOT ported. e2e/public.sh publishes 25, 80, 443,
 # 465, 993, 995 and 4190 on the machine it runs on, which on GitHub was a
 # throwaway VM and here would be the CI host -- where 80 and 443 are nginx
 # and the mail ports belong to the mail netns. It stays a manual check on a
 # disposable host until there is a runner that can safely be given those
 # ports.
 name: ci
 on:
  push:
    branches: [main]
    tags: ["v*"]
  pull_request:
 concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: true
 jobs:
  test:
    runs-on: docker
    container:
      image: golang:1.26-bookworm@sha256:a688600ca24f8a4d3ca77f95b0dd40704a9fc787c826660eb7ba0b641b8b175d # 1.26-bookworm
    steps:
      - uses: coffey-labs/actions/checkout@fab0c4d45e0162963965f1555df27b7bed5e20ec
      - run: go vet ./...
      - run: go test ./...
      # Kept as `go run ...@latest` exactly as the workflow had it: the point
      # of a vulnerability check is to use today's database, not a pinned copy
      # of last month's.
      - run: go run golang.org/x/vuln/cmd/govulncheck@latest ./...
  release:
    if: startsWith(github.ref, 'refs/tags/')
    needs: [test]
    runs-on: docker
    container:
      image: golang:1.26-bookworm@sha256:a688600ca24f8a4d3ca77f95b0dd40704a9fc787c826660eb7ba0b641b8b175d # 1.26-bookworm
    steps:
      # Full history: the ancestry check below cannot be answered from a
      # shallow clone. The checkout also fetches every branch as origin/*.
      - uses: coffey-labs/actions/checkout@fab0c4d45e0162963965f1555df27b7bed5e20ec
        with:
          fetch-depth: 0
      # The workflow refused to release a tag that is not an ancestor of main,
      # so that a release can never describe code that was never reviewed onto
      # the default branch.
      - shell: bash
        env:
          TAG: ${{ github.ref_name }}
        run: |
          git merge-base --is-ancestor "$(git rev-parse "${TAG}^{commit}")" origin/main \
            || { echo "!! $TAG is not on main"; exit 1; }
      # SOURCE_DATE_EPOCH is what makes the tarballs reproducible: without it
      # every build stamps a new mtime and two builds of one tag differ.
      - shell: bash
        env:
          TAG: ${{ github.ref_name }}
        run: |
          SOURCE_DATE_EPOCH="$(git log -1 --format=%ct "$TAG")" scripts/build-release.sh "$TAG" dist
          sha256sum dist/*.tar.gz
      # Create the Release, then attach every file. Archive names carry no
      # version, so /releases/latest/download/<name> always means the newest.
      # The API is reached on the internal address so the uploads never cross
      # Cloudflare. If an upload fails the half-made Release is deleted: a
      # Release whose assets 404 is worse than no Release, since the install
      # guide sends people straight at these URLs.
      - shell: bash
        env:
          TAG: ${{ github.ref_name }}
          TOKEN: ${{ secrets.GITHUB_TOKEN }}
          REPO: ${{ github.repository }}
        run: |
          set -euo pipefail
          # CI_SERVER_INTERNAL is set on every job container by the runner.
          API="$CI_SERVER_INTERNAL/api/v1/repos/$REPO"
          auth=(--header "Authorization: token $TOKEN")
          id=$(curl --fail --silent --show-error "${auth[@]}" \
                --header "Content-Type: application/json" \
                --data "{\"tag_name\":\"$TAG\",\"name\":\"$TAG\",\"body\":\"Binaries for linux/amd64 and linux/arm64. Verify with SHA256SUMS.\"}" \
                "$API/releases" | grep -o '^{"id":[0-9]*' | cut -d: -f2)
          [ -n "$id" ] || { echo "!! could not create the release"; exit 1; }
          for f in dist/*; do
            n=$(basename "$f")
            echo "uploading $n"
            curl --fail --silent --show-error --output /dev/null "${auth[@]}" \
              --form "attachment=@$f" "$API/releases/$id/assets?name=$n" \
              || { curl --silent "${auth[@]}" -X DELETE "$API/releases/$id"; exit 1; }
          done
@@ -0,0 +1,41 @@
 # The end-to-end test, every week, against ihasmail's newest release.
 #
 # A deploy with no --ihasmail-image takes whatever ihasmail release is newest
 # when it runs, rather than one this repository pinned. That is only safe if
 # something notices when a new release stops working with the Stalwart and
 # Caddy versions pinned here -- this is that something. It runs on Mondays
 # three hours after ihasmail's weekly release (09:00 UTC), so a release that
 # breaks the one shot shows up the same day, before most people deploy it.
 #
 # e2e/public.sh needs nothing from the internet but images: Pebble stands in for
 # Let's Encrypt and a DNS stub answers every name, and it publishes 25, 80, 443,
 # 465, 993, 995 and 4190 on the runner while it runs.
 name: End-to-end
 on:
  schedule:
    - cron: "0 12 * * 1"
  workflow_dispatch:
 concurrency:
  group: e2e
  cancel-in-progress: false
 permissions:
  contents: read
 jobs:
  public:
    runs-on: ubuntu-latest
    timeout-minutes: 30
    steps:
      - uses: actions/checkout@v7
      - uses: actions/setup-go@v7
        with:
          go-version-file: go.mod
      - name: Vet and test
        run: |
          go vet ./...
          go test ./...
      - name: Deploy and check a public stack
        run: e2e/public.sh
@@ -0,0 +1,116 @@
 # CI on the self-hosted GitLab, ported from .github/workflows/release.yml when
 # the GitHub account was suspended on 2026-09-20. The Actions file stays in the
 # tree: it is the reference this was written from and works unchanged if the
 # appeal succeeds.
 #
 # e2e.yml is deliberately NOT ported. e2e/public.sh publishes 25, 80, 443,
 # 465, 993, 995 and 4190 on the machine it runs on, which on GitHub was a
 # throwaway VM and here would be Web_Host -- where 80 and 443 are nginx
 # serving every live site and the mail ports belong to the mail netns.
 # Running it on this runner would take the sites down for the length of the
 # test. It stays a manual check on a disposable host until there is a runner
 # that can safely be given those ports.
 #
 # The shape is the same -- tag-driven, amd64 and arm64, reproducible -- but the
 # publishing half is necessarily different. There is no `gh release`, so the
 # tarballs go to this project's generic package registry and the Release is
 # created with release-cli, linking to them. The docs guide installs from
 # release assets, so those links are the part that has to keep working.
 #
 # Images are pinned by digest, with the tag in the trailing comment: the
 # replacement for the workflow's SHA-pinned actions, since GitLab has no
 # action allowlist.
 stages: [test, build, release]
 variables:
  PKG: "${CI_API_V4_URL}/projects/${CI_PROJECT_ID}/packages/generic/ihasmail-oneshot"
 default:
  interruptible: true
 .go: &go
  image: golang:1.26-bookworm@sha256:a688600ca24f8a4d3ca77f95b0dd40704a9fc787c826660eb7ba0b641b8b175d # 1.26-bookworm
  cache:
    key: go-mod
    paths: [.gocache/]
  variables:
    GOPATH: "$CI_PROJECT_DIR/.gocache"
 test:
  <<: *go
  stage: test
  script:
    - go vet ./...
    - go test ./...
    # Kept as `go run ...@latest` exactly as the workflow had it: the point of
    # a vulnerability check is to use today's database, not a pinned copy of
    # last month's.
    - go run golang.org/x/vuln/cmd/govulncheck@latest ./...
  rules:
    - if: $CI_PIPELINE_SOURCE == "merge_request_event"
    - if: $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH
    - if: $CI_COMMIT_TAG
 build:
  <<: *go
  stage: build
  needs: [test]
  script:
    # The workflow refused to release a tag that is not an ancestor of main,
    # so that a release can never describe code that was never reviewed onto
    # the default branch. GIT_DEPTH is unset below to make the ancestry
    # available -- a shallow clone cannot answer this.
    - git fetch --quiet origin "$CI_DEFAULT_BRANCH"
    - |
      git merge-base --is-ancestor "$(git rev-parse "${CI_COMMIT_TAG}^{commit}")" "origin/$CI_DEFAULT_BRANCH" \
        || { echo "!! $CI_COMMIT_TAG is not on $CI_DEFAULT_BRANCH"; exit 1; }
    # SOURCE_DATE_EPOCH is what makes the tarballs reproducible: without it
    # every build stamps a new mtime and two builds of one tag differ.
    - SOURCE_DATE_EPOCH="$(git log -1 --format=%ct "$CI_COMMIT_TAG")" scripts/build-release.sh "$CI_COMMIT_TAG" dist
    - sha256sum dist/*.tar.gz
  variables:
    GIT_DEPTH: "0"
  artifacts:
    paths: [dist/]
    expire_in: 1 week
  rules:
    - if: $CI_COMMIT_TAG
 release:
  stage: release
  image: registry.gitlab.com/gitlab-org/cli:latest@sha256:3f0a591b3b96c39ac8e28480ee99bb93201b7bcea1fbca7ed50c034098111db2 # latest
  needs: [build]
  script:
    # Upload first, then create the Release pointing at what was uploaded. A
    # Release whose assets 404 is worse than no Release: the install guide
    # sends people straight at these URLs.
    - |
      set -eu
      for f in dist/*; do
        n=$(basename "$f")
        echo "uploading $n"
        curl --fail --silent --show-error \
          --header "JOB-TOKEN: ${CI_JOB_TOKEN}" \
          --upload-file "$f" \
          "${PKG}/${CI_COMMIT_TAG}/${n}"
      done
    - |
      set -eu
      args=""
      for f in dist/*; do
        n=$(basename "$f")
        # direct_asset_path is what makes the permalink work. Without it the
        # only stable URL is the package registry one, which carries the tag
        # and so cannot be a "latest" link. With it, every release exposes
        #   /-/releases/permalink/latest/downloads/binaries/<file>
        # which is the GitLab equivalent of the GitHub
        # /releases/latest/download/<file> URL the install guide has always
        # used. Changing this path breaks documented install commands.
        args="$args --assets-link {\"name\":\"${n}\",\"url\":\"${PKG}/${CI_COMMIT_TAG}/${n}\",\"direct_asset_path\":\"/binaries/${n}\"}"
      done
      # shellcheck disable=SC2086
      release-cli create --name "$CI_COMMIT_TAG" --tag-name "$CI_COMMIT_TAG" \
        --description "Binaries for linux/amd64 and linux/arm64. Verify with SHA256SUMS." $args
  rules:
    - if: $CI_COMMIT_TAG
@@ -0,0 +1,81 @@
 # Contributing to ihasmail-oneshot
 How to build and test the tool, how the code is organized, and how versions and
 releases work. Back to the [README](README.md).
 ## Building
 With Go 1.26.8 or newer:
 ```bash
 go build -o ihasmail-oneshot ./cmd/ihasmail-oneshot
 ```
 ## Testing
 ```bash
 go vet ./...
 go test ./...          # unit tests: validation, rendered files, the Stalwart client
 e2e/public.sh          # the whole mail-host deploy, for real, on this machine
 ```
 `e2e/public.sh` runs a complete mail-host deployment with no internet involved.
 [Pebble](https://github.com/letsencrypt/pebble), the ACME test server, stands in
 for Let's Encrypt, and a DNS stub answers every name with the host's own address.
 Caddy and Stalwart both obtain real certificates from it, through the real
 ports and the real `Caddyfile`. It then checks, among other things, that:
 - the deploy completes, links ihasmail and Stalwart, and reports both kinds of
  certificate;
 - `credentials.txt` and `.env` are private, and the bootstrap credential is gone
  from the Stalwart container;
 - Stalwart's IMAPS (993) and submission (465) ports present certificates that
  verify for the mail host;
 - the webmail, Stalwart's JMAP and autoconfig are served over verified HTTPS, and
  a user signs in through the webmail;
 - `dns-records.zone` holds the MX and DKIM records, and `certs` recognizes an
  existing certificate;
 - a scanner probing through Caddy is banned by its own address, not Caddy's, and
  other clients still get through;
 - ihasmail is exempt from bans, and a user still signs in after failed attempts.
 It publishes ports 25, 80, 443, 465, 993, 995 and 4190 on the machine while it
 runs, and removes everything it created when it ends, pass or fail. `KEEP=1
 e2e/public.sh` leaves the stack up to inspect.
 ## How the code is laid out
 | Package | Does |
 | --- | --- |
 | `cmd/ihasmail-oneshot` | Commands, flags, confirmation, the summary |
 | `internal/config` | Validates the command line into a plan: names, addresses, images |
 | `internal/render` | Renders `compose.yaml` and the `Caddyfile`; writes files without ever overwriting |
 | `internal/docker` | Drives the `docker` and `docker compose` CLIs |
 | `internal/stalwart` | The JMAP client and every Stalwart registry call: bootstrap, ACME, bans, mailboxes, DNS zone |
 | `internal/webmail` | ihasmail's health check and the sign-in that proves the link |
 | `internal/deploy` | Preflight, the deploy sequence, `certs`, `destroy` |
 ## Versions and releases
 Releases are tagged by date, like ihasmail's: `v2026.9.13`, with `.1`, `.2`
 added for another release the same day. Each release pins the Stalwart and
 Caddy images it was tested with as its defaults. A newer release of the tool
 generally means newer tested versions of those.
 ihasmail is the exception: a deploy takes its newest release, so a new
 ihasmail needs no new release of this tool. What keeps that safe is the
 end-to-end test, which runs every Monday against ihasmail's newest release, a
 few hours after ihasmail publishes it. Stalwart is never taken this way — an
 upgrade migrates its data with no way back, so its version only changes in a
 release of this tool.
 Binaries for `linux/amd64` and `linux/arm64` and a `SHA256SUMS` file are
 attached to every [release](https://github.com/Coffey-Labs/ihasmail-oneshot/releases).
 Every release is built by the [release workflow](.github/workflows/release.yml)
 from a tagged commit on `main`, after the tests and a known-vulnerabilities
 check pass. The archives are reproducible: `scripts/build-release.sh` builds the
 same bytes from the same commit.
 That weekly run is [`e2e.yml`](.github/workflows/e2e.yml), Mondays at 12:00
 UTC. It can also be started by hand from the Actions tab.
@@ -2,6 +2,7 @@
 [![Latest release](https://img.shields.io/github/v/release/Coffey-Labs/ihasmail-oneshot?sort=date)](https://github.com/Coffey-Labs/ihasmail-oneshot/releases/latest)
 [![License: AGPL-3.0-or-later](https://img.shields.io/badge/license-AGPL--3.0--or--later-blue)](LICENSE)
 [![Docs: docs.ihasmail.org](https://img.shields.io/badge/docs-docs.ihasmail.org-0ea5e9)](https://docs.ihasmail.org/install/oneshot/)
 **One command that turns a Linux Docker host into a working mail server with
 webmail.** It deploys a fresh [Stalwart](https://stalw.art) mail server and a
@@ -13,158 +14,39 @@ the DNS records to publish.
 ihasmail-oneshot deploy --domain example.com --user alice
 ```
-When it finishes you have a mail server for `example.com`, webmail at
+## Documentation
 `https://webmail.example.com`, Stalwart's admin UI at
 `https://mail.example.com/admin`, and a mailbox for `[email protected]`. Once
 you publish the DNS records it hands you, mail flows in and out. What it
 leaves behind is an ordinary `docker compose` project, managed with the usual
 commands and not tied to this tool.
---
+| | |
 | --- | --- |
 | 📘 **[Step-by-step guide](https://docs.ihasmail.org/install/oneshot/)** | **Start here.** On docs.ihasmail.org: DNS, ports, deploying, publishing records, signing in, upgrading and backups |
 | 📋 **[Reference](docs/reference.md)** | Requirements, every command and flag, the deployment directory, upgrading, backing up, removing |
 | ⚙️ **[How it works](docs/how-it-works.md)** | Why it exists, the deploy sequence, Stalwart setup without its wizard, shared certificates, IP bans behind a proxy |
 | 🔒 **[Security model](docs/security-model.md)** | What is exposed, where secrets live, the trust decisions it makes |
 | 🧰 **[Troubleshooting](docs/troubleshooting.md)** | Problems by message or symptom, and known limits |
 | 🧪 **[Contributing](CONTRIBUTING.md)** | Building, the unit and end-to-end tests, the code layout, releases |
- [What it is](#what-it-is)
+## What you get
 - [What it does](#what-it-does)
 - [Why it exists](#why-it-exists)
 - [Requirements](#requirements)
 - [Installing](#installing)
 - [Trying it locally in two minutes](#trying-it-locally-in-two-minutes)
 - [Deploying a mail host, step by step](#deploying-a-mail-host-step-by-step)
 - [The deployment directory](#the-deployment-directory)
 - [Running it afterwards](#running-it-afterwards)
 - [Command reference](#command-reference)
 - [How it works](#how-it-works)
 - [Security model](#security-model)
 - [Troubleshooting](#troubleshooting)
 - [Known limits](#known-limits)
 - [Development and testing](#development-and-testing)
 - [Versions and releases](#versions-and-releases)
 - [License](#license)
---
+- A mail server for `example.com`, with webmail at `https://webmail.example.com`
-
+  and Stalwart's admin UI at `https://mail.example.com/admin`.
-## What it is
+- A mailbox for each `--user`, and an administrator, with generated passwords in
-
+  `credentials.txt`.
-A single, statically linked binary for Linux (amd64 and arm64). You run it on
+- Certificates for the webmail and for Stalwart's mail ports.
-the Docker host that should become the mail server. It needs Docker and the
+- `dns-records.zone`: every DNS record to publish, DKIM keys included.
-compose plugin, and nothing else: no configuration file to write first, no
+- An ordinary `docker compose` project, managed with the usual commands and not
-language runtime, no other scripts.
+  tied to this tool. ihasmail runs read-only with no volume; everything durable
-
+  is in Stalwart's volume.
 It deploys three containers:
 | Container | Image | Role |
 | --- | --- | --- |
 | **Stalwart** | `stalwartlabs/stalwart` | The mail server: SMTP, IMAP, POP3, JMAP, CalDAV, CardDAV, spam filtering, DKIM. It holds all the mail and all the accounts |
 | **ihasmail** | `ghcr.io/coffey-labs/ihasmail` | The webmail: mail, calendars, contacts, files and filters in the browser, talking to Stalwart over JMAP. It holds nothing but sessions |
 | **Caddy** | `caddy` | The HTTPS front: certificates and TLS for the webmail and for Stalwart's web side |
 It has two shapes:
 | | `deploy --domain example.com` | `deploy --local` |
 | --- | --- | --- |
 | **For** | A real mail host on the internet | Trying ihasmail against a real Stalwart on your own machine |
 | **Containers** | Stalwart, ihasmail, Caddy | Stalwart, ihasmail |
 | **Publishes** | 25, 80, 443, 465, 993, 995, 4190 on every interface | Nothing but two loopback ports |
 | **Certificates** | Let's Encrypt, for both Caddy and Stalwart | None |
 | **Webmail** | `https://webmail.example.com` | `http://127.0.0.1:8080` |
 | **Stalwart admin** | `https://mail.example.com/admin` | `http://127.0.0.1:8081/admin` |
 ## What it does
 In order, in one run:
 1. **Checks the host** before changing anything: Docker and compose are there
   and usable, every port it needs is free, the deployment directory is new or
   empty, no compose project of the same name exists, and the hostnames resolve.
   It reports every problem at once, not the first.
 2. **Shows you the plan and asks** before going ahead. `--yes` skips the
   question, and is required when there is no terminal to ask on.
 3. **Writes the deployment directory**: `compose.yaml`, the `Caddyfile`, and
   `.env` holding a freshly generated `APP_SECRET`.
 4. **Pulls the images**, pinned to versions tested together.
 5. **Starts Stalwart in bootstrap mode** with a one-time administrator whose
   password exists only in the tool's memory.
 6. **Completes Stalwart's setup** through its API, the same setup its web UI
   would otherwise walk you through: hostname, mail domain, DKIM keys, logging.
   Stalwart creates the permanent administrator and returns its password, which
   is written to `credentials.txt` straight away.
 7. **Starts the whole stack.** Stalwart is recreated *without* the one-time
   administrator.
 8. **Links ihasmail and Stalwart**: exempts ihasmail from Stalwart's automatic
   IP bans, tells Stalwart to trust the client addresses Caddy forwards, and
   restarts Stalwart so both apply.
 9. **Creates the mailboxes** you asked for with `--user`, each with a generated
   password.
 10. **Proves the link** by signing in *through ihasmail* as the administrator.
    That only succeeds if ihasmail can reach Stalwart and Stalwart accepts the
    credentials.
 11. **Requests certificates** (mail host only): sets up an ACME account in
    Stalwart for its IMAP and SMTP certificate, and waits for that and for
    Caddy's certificates to arrive.
 12. **Writes `dns-records.zone`**, every DNS record Stalwart wants published,
    and prints a summary of URLs, credentials and anything still to do.
 ## Why it exists
 Stalwart and ihasmail each install easily on their own. Making them a working,
 safe mail host *together* takes a series of decisions that are easy to get
 wrong, and most of the mistakes don't show until later:
 - **Stalwart 0.16 has no configuration file to template.** Its settings live in
  its data store, and a new server starts in a bootstrap mode that expects a
  person at a web wizard. Automating that means driving its registry API, and
  making sure the bootstrap credential doesn't quietly outlive the setup.
 - **Both services want the same ports for certificates.** The webmail needs
  HTTPS on 443. Stalwart needs its own certificate for IMAP and SMTP, and its
  default way of getting one also wants port 443. Two ACME clients fighting over
  one port fail in confusing, intermittent ways.
 - **Stalwart's automatic IP bans don't expect a proxy in front of it.** Stalwart
  bans addresses that probe for things like WordPress admin pages. Behind a
  reverse proxy, every client arrives from the proxy's address, so one bot
  scanning your mail host gets *the proxy* banned. Autoconfig, calendar sync and
  certificate renewals all stop working, for everyone, with nothing obviously
  wrong. This was reproduced on Stalwart 0.16.22 while building this tool, as
  was a second trap: the setting that fixes it only takes effect after a
  restart.
 - **The route from webmail to mail server matters.** Sent over the private
  Docker network, it never leaves the host and costs about a third of the
  memory per signed-in browser tab that going back out over public HTTPS does.
 - **Mail doesn't flow until DNS is right**, and the list of records (MX, SPF,
  DKIM, DMARC, SRV, MTA-STS, autoconfig) is long.
 This tool makes each of those decisions once, the same way every time, and
 checks the result before telling you it is done. Each one is explained in
 [How it works](#how-it-works).
 ## Requirements
-**On the host:**
+- Linux (amd64 or arm64) with Docker Engine and the compose plugin.
 - For a mail host: a domain whose DNS you control, a static public IP, ports
  **25, 80, 443, 465, 993, 995 and 4190** open, **outbound port 25** allowed by
  your provider, and **reverse DNS** for the host's address naming the mail host.
- Linux, amd64 or arm64.
+The full list is in [docs/reference.md](docs/reference.md#requirements).
 - Docker Engine with the compose plugin (`docker compose version` works), run
  as a user allowed to use Docker.
-**For a mail host, additionally:**
+## Install
 - A domain whose DNS you control.
 - **Ports 25, 80, 443, 465, 993, 995 and 4190** free on the host and open in any
  firewall or cloud security group in front of it.
 - **Outbound port 25.** Many cloud and VPS providers block it by default and
  unblock it on request. Without it you can receive mail but not send it.
 - **Reverse DNS**: a PTR record for the host's IP address naming the mail host
  (`mail.example.com`). This is set at your hosting provider, not in your DNS
  zone. Many receiving servers reject mail from an address without one.
 - A static public IP address.
 **Resources:** ihasmail on its own needs 256 MiB of RAM at minimum. Stalwart's
 needs grow with the mail it stores and the number of people using it; size the
 host for Stalwart.
 ## Installing
 ### From a release (recommended)
 Download the archive for your architecture and its checksums, verify, and
 unpack:
 ```bash
 ARCH=amd64   # or arm64
@@ -173,694 +55,39 @@ curl -fsSLO https://github.com/Coffey-Labs/ihasmail-oneshot/releases/latest/down
 sha256sum --ignore-missing -c SHA256SUMS
 tar -xzf ihasmail-oneshot-linux-$ARCH.tar.gz
 sudo install -m 0755 ihasmail-oneshot /usr/local/bin/
 ihasmail-oneshot version
 ```
-Every release is built by the [release workflow](.github/workflows/release.yml)
+Or build from source with Go 1.26.8 or newer:
-from a tagged commit on `main`, after the tests and a known-vulnerabilities
+`go build -o ihasmail-oneshot ./cmd/ihasmail-oneshot`.
 check pass. The archives are reproducible: `scripts/build-release.sh` builds the
 same bytes from the same commit.
-### From source
+## Use
-With Go 1.26.8 or newer:
+Try ihasmail against a real Stalwart on your own machine, with no domain and
-
+nothing reachable from outside:
 ```bash
 git clone https://github.com/Coffey-Labs/ihasmail-oneshot.git
 cd ihasmail-oneshot
 go build -o ihasmail-oneshot ./cmd/ihasmail-oneshot
 ```
 ## Trying it locally in two minutes
 To see ihasmail against a real Stalwart on your own machine, without a domain
 or open ports:
 ```bash
 ihasmail-oneshot deploy --local --user alice
 ihasmail-oneshot destroy --dir ihasmail-example-test     # when you're done
 ```
-```text
+Deploy a real mail host, after pointing `mail.example.com` and
-==> a local pair for example.test, in /home/you/ihasmail-example-test
+`webmail.example.com` at it:
    ihasmail  http://127.0.0.1:8080
    Stalwart  http://127.0.0.1:8081  (admin UI; no mail ports published)
    images    stalwartlabs/stalwart:v0.16.22, ghcr.io/coffey-labs/ihasmail:2026.9.10-pr328
    mailboxes alice
 ==> preflight
    docker 29.8.0, compose 5.5.1
 deploy this? [y/N] y
 ==> writing /home/you/ihasmail-example-test
 ==> pulling images
 ==> starting Stalwart in bootstrap mode
 ==> setting up Stalwart for example.test (hostname mail.example.test)
    administrator [email protected], password in credentials.txt
 ==> starting the stack
 ==> linking ihasmail and Stalwart
    ihasmail (172.31.253.11) exempt from Stalwart's auto-ban
    restarting Stalwart to apply them
    mailbox [email protected] created
    signed in to ihasmail 2026.9.10+pr328 as [email protected]: linked
 ==> done
    webmail      http://127.0.0.1:8080
    Stalwart     http://127.0.0.1:8081/admin
    sign in as   [email protected] (password in /home/you/ihasmail-example-test/credentials.txt)
                 the administrator gets the webmail's Administration menu
    mailbox      [email protected]
 ```
 Open `http://127.0.0.1:8080` and sign in as `[email protected]` with the
 password from `ihasmail-example-test/credentials.txt`. Nothing outside your
 machine can reach the pair, and no mail from outside can be delivered to it.
 When you're done:
 ```bash
-ihasmail-oneshot destroy --dir ihasmail-example-test
+ihasmail-oneshot deploy --domain example.com --email [email protected] --user alice
 ```
-## Deploying a mail host, step by step
+It checks the host, shows the plan with the exact ihasmail release it will use,
 and asks before changing anything. Then publish `dns-records.zone`, and mail
 flows.
-The example uses `example.com`. Substitute your own domain throughout.
+## Versions
-### 1. Point two names at the host
+Releases are tagged by date (`v2026.9.15`). Each pins the Stalwart and Caddy
-
+versions it was tested with. ihasmail is the newest release at deploy time,
-Before running anything, create these records at your DNS provider, using the
+written into `compose.yaml` by its dated tag, and checked by an end-to-end test
-host's public addresses:
+every Monday. Nothing upgrades by itself afterwards. Details in
-
+[CONTRIBUTING.md](CONTRIBUTING.md#versions-and-releases).
 ```dns
 mail.example.com.     IN A     203.0.113.10
 webmail.example.com.  IN A     203.0.113.10
 ; and, if the host has IPv6:
 mail.example.com.     IN AAAA  2001:db8::10
 webmail.example.com.  IN AAAA  2001:db8::10
 ```
 Doing this first means certificates are issued during the deploy. If you skip
 it, the deploy still completes and tells you how to finish once DNS is in place.
 > If your DNS provider offers a proxy or CDN mode, leave these records
 > **DNS-only**. Mail ports cannot be proxied, and certificate validation must
 > reach this host directly.
 ### 2. Open the ports
 Allow inbound TCP 25, 80, 443, 465, 993, 995 and 4190, and UDP 443 (HTTP/3),
 in the host firewall and in any provider firewall. Ask your provider to unblock
 **outbound** port 25 if they block it, and set **reverse DNS** for the host's
 address to `mail.example.com`.
 ### 3. Run the deploy
 On the host:
 ```bash
 ihasmail-oneshot deploy \
  --domain example.com \
  --email [email protected] \
  --user alice --user bob
 ```
 `--email` is the address Let's Encrypt associates with your certificates. Use
 one that doesn't depend on this new server working. It defaults to
 `[email protected]`.
 The tool prints its plan, checks the host, and asks for confirmation. A run
 takes a minute or two, most of it pulling images. It ends with a summary:
 ```text
 ==> done
    webmail      https://webmail.example.com
    Stalwart     https://mail.example.com/admin
    sign in as   [email protected] (password in /root/ihasmail-example-com/credentials.txt)
                 the administrator gets the webmail's Administration menu
    mailbox      [email protected]
    mailbox      [email protected]
    DNS          /root/ihasmail-example-com/dns-records.zone -- publish every record in it
    certificate  issued by <Let's Encrypt intermediate>, valid until <date>
    https        webmail.example.com: issued by <Let's Encrypt intermediate>
    https        mail.example.com: issued by <Let's Encrypt intermediate>
    Before mail flows: reverse DNS for this host's address should name mail.example.com,
    and outbound port 25 must be open -- many providers block it until asked.
 ```
 The last three lines are how you know certificates were issued. If any is
 replaced by a warning, see [step 5](#5-get-stalwarts-certificate-if-dns-came-late).
 ### 4. Publish the DNS records
 `dns-records.zone` holds every record Stalwart wants published, including the
 DKIM keys it just generated:
 ```dns
 v1-ed25519-20260913._domainkey.example.com. IN TXT "v=DKIM1; k=ed25519; h=sha256; p=…"
 v1-rsa-20260913._domainkey.example.com. IN TXT ( "v=DKIM1; k=rsa; h=sha256; p=…" )
 mail.example.com. IN TXT "v=spf1 a -all"
 example.com. IN TXT "v=spf1 mx -all"
 example.com. IN MX 10 mail.example.com.
 _dmarc.example.com. IN TXT "v=DMARC1; p=reject; rua=mailto:[email protected]"
 _caldavs._tcp.example.com. IN SRV 0 1 443 mail.example.com.
 _carddavs._tcp.example.com. IN SRV 0 1 443 mail.example.com.
 _imaps._tcp.example.com. IN SRV 0 1 993 mail.example.com.
 _jmap._tcp.example.com. IN SRV 0 1 443 mail.example.com.
 _pop3s._tcp.example.com. IN SRV 0 1 995 mail.example.com.
 _submissions._tcp.example.com. IN SRV 0 1 465 mail.example.com.
 mta-sts.example.com. IN CNAME mail.example.com.
 _mta-sts.example.com. IN TXT "v=STSv1; id=…"
 _smtp._tls.example.com. IN TXT "v=TLSRPTv1; rua=mailto:[email protected]"
 autoconfig.example.com. IN CNAME mail.example.com.
 autodiscover.example.com. IN CNAME mail.example.com.
 …
 ```
 Publish all of them. Many DNS providers can import a zone file directly. The
 **MX**, **SPF**, **DKIM** and **DMARC** records decide whether your mail is
 delivered and whether others' mail reaches you. The rest let mail apps configure
 themselves from just an email address.
 > DMARC is published as `p=reject`: receivers are told to reject mail that fails
 > SPF and DKIM. That's the right policy for a domain that sends only from this
 > server. If other services send mail as your domain, start with `p=none`.
 ### 5. Get Stalwart's certificate, if DNS came late
 If the names didn't resolve during the deploy, the summary says Stalwart has no
 certificate yet. Caddy keeps retrying its own certificates, but Stalwart doesn't
 retry a failed order. Once DNS points at the host:
 ```bash
 ihasmail-oneshot certs --dir ihasmail-example-com
 ```
 This works now that every Stalwart name has a record, including the
 `autoconfig`, `autodiscover`, `mta-sts` and `ua-auto-config` CNAMEs from
 `dns-records.zone`, because Stalwart's certificate covers all of them.
 ### 6. Sign in
 - **Webmail:** `https://webmail.example.com`, as any mailbox. As
  `[email protected]` you also get the Administration menu, for adding people
  and domains.
 - **Stalwart's admin UI:** `https://mail.example.com/admin`, as
  `[email protected]`, for every server setting.
 - **Mail apps** (Thunderbird, Apple Mail, phones): add the account by email
  address and password. Once the autoconfig records are published, most apps
  find the servers themselves. By hand, they're IMAP on `mail.example.com:993`
  (TLS) and SMTP submission on `mail.example.com:465` (TLS).
 **Change the generated passwords** after first sign-in, then delete them from
 `credentials.txt` or keep that file somewhere safe. An account with two-factor
 authentication turned on signs in to the webmail with an app password created
 in Stalwart.
 ## The deployment directory
 The deploy writes a directory named after the domain, `./ihasmail-example-com`
 by default (`--dir` to choose):
 | File | Mode | What it is |
 | --- | --- | --- |
 | `compose.yaml` | 0644 | The whole deployment: three services, a private network, named volumes. Commented |
 | `Caddyfile` | 0644 | Caddy's configuration: the webmail, Stalwart's web names, and port 80's challenge forwarding |
 | `.env` | 0600 | `APP_SECRET`, which seals ihasmail's session cookies. Read by compose |
 | `credentials.txt` | 0600 | The administrator's and each mailbox's generated password, plus what `certs` needs to reach Stalwart |
 | `dns-records.zone` | 0644 | The DNS records to publish |
 | `acme-ca-root.pem`, `ca-bundle.crt` | 0644 | Only with `--acme-ca-root`: the private CA, and the system roots with it added |
 Data lives in Docker named volumes, prefixed with the project name:
 | Volume | Holds |
 | --- | --- |
 | `…_stalwart-data` | All mail, accounts, calendars, contacts, files, settings, DKIM keys, Stalwart's certificates |
 | `…_stalwart-etc` | Stalwart's store location file |
 | `…_caddy-data` | Caddy's certificates and ACME account |
 | `…_caddy-config` | Caddy's autosaved configuration |
 ihasmail has no volume. It runs read-only and keeps sessions in memory.
 ## Running it afterwards
 The directory is a standard compose project. From inside it:
 ```bash
 docker compose ps                      # what's running
 docker compose logs -f stalwart        # follow Stalwart's log
 docker compose logs caddy | grep -i error
 docker compose restart ihasmail        # signs everyone out of the webmail; nothing else is lost
 docker compose down                    # stop everything (data stays in the volumes)
 docker compose up -d                   # start it again
 ```
 The containers restart by themselves after a crash or a reboot
 (`restart: unless-stopped`).
 ### Upgrading
 Change the image tag in `compose.yaml` and apply it:
 ```bash
 docker compose pull && docker compose up -d
 ```
 - **ihasmail** is safe to move to any newer release that supports your
  Stalwart version. Its release notes say which.
 - **Stalwart**: read its upgrade notes before changing versions. Check that the
  ihasmail version you run supports the new Stalwart release first, since
  ihasmail validates against one Stalwart release at a time. Back up
  `stalwart-data` first.
 - **Caddy** minor releases are routine.
 ### Backing up
 Everything that can't be recreated is in the `stalwart-data` volume. For a
 consistent copy, stop Stalwart briefly:
 ```bash
 docker compose stop stalwart
 docker run --rm -v ihasmail-example-com_stalwart-data:/data -v "$PWD":/backup alpine \
  tar -czf /backup/stalwart-data-$(date +%F).tar.gz -C /data .
 docker compose start stalwart
 ```
 Keep `caddy-data` too, or certificates are requested again on a rebuild. That's
 harmless unless it happens often enough to meet Let's Encrypt's rate limits. Keep
 the deployment directory itself, which is small and contains your secrets.
 ### Removing a deployment
 ```bash
 ihasmail-oneshot destroy --dir ihasmail-example-com
 ```
 This removes the containers, the network, **the volumes with every message and
 account**, and the files the tool wrote. It asks first unless given `--yes`.
 Files you added to the directory yourself are left in place, and so is the
 directory.
 ## Command reference
 ### `deploy`
 ```text
 ihasmail-oneshot deploy --domain example.com [flags]
 ihasmail-oneshot deploy --local [flags]
 ```
 | Flag | Default | Meaning |
 | --- | --- | --- |
 | `--domain` | required, or `example.test` with `--local` | The mail domain this server receives for |
 | `--local` | off | The loopback-only shape: no mail ports, no Caddy, no certificates |
 | `--mail-host` | `mail.DOMAIN` | Stalwart's hostname. Must be one label under the domain, e.g. `mx.example.com` |
 | `--webmail-host` | `webmail.DOMAIN` | The webmail's hostname. Any name except Stalwart's |
 | `--email` | `postmaster@DOMAIN` | ACME contact address for both Caddy and Stalwart |
 | `--user NAME` | none | Create a mailbox `NAME@DOMAIN` with a generated password. Repeat for more |
 | `--dir` | `./PROJECT` | Deployment directory to write. Must be new or empty |
 | `--project` | `ihasmail-DOMAIN` (dots as dashes) | Compose project name, which prefixes containers, network and volumes |
 | `--stalwart-image` | `stalwartlabs/stalwart:v0.16.22` | Stalwart image |
 | `--ihasmail-image` | `ghcr.io/coffey-labs/ihasmail:2026.9.10-pr328` | ihasmail image |
 | `--caddy-image` | `caddy:2.11.4` | Caddy image |
 | `--webmail-bind` | `127.0.0.1:8080` | Host address for ihasmail's own port, for reaching it without Caddy |
 | `--stalwart-bind` | `127.0.0.1:8081` | Host address for Stalwart's plain-HTTP port. The tool configures Stalwart through it |
 | `--subnet` | `172.31.253.0/24` | The stack's private network. Change it if it overlaps a network you already have |
 | `--acme-directory` | Let's Encrypt | ACME directory URL of a private CA, for both Caddy and Stalwart |
 | `--acme-ca-root` | none | PEM file of the root the private CA's HTTPS endpoint is signed by |
 | `--yes` | off | Don't ask for confirmation. Required without a terminal |
 The defaults are the versions tested together for this release.
 ### `certs`
 ```text
 ihasmail-oneshot certs --dir DIR
 ```
 Starts a new certificate order in Stalwart for the deployment in `DIR` and waits
 for it, typically after fixing DNS. If Stalwart already holds a valid
 certificate for the mail host, it reports that and does nothing.
 ### `destroy`
 ```text
 ihasmail-oneshot destroy --dir DIR [--yes]
 ```
 Removes the deployment in `DIR` with all of its data. See
 [Removing a deployment](#removing-a-deployment).
 ### `version`
 Prints the tool's version.
 ## How it works
 ### The shape of the deployment
 ```mermaid
 flowchart LR
    Browser(["Browser"])
    Apps(["Mail apps, other mail servers"])
    subgraph Host["Docker host"]
        subgraph Net["private network 172.31.253.0/24"]
            Caddy["Caddy<br/>.10"]
            Ihasmail["ihasmail<br/>.11<br/>read-only, no volume"]
            Stalwart["Stalwart<br/>.12"]
        end
        Data[("stalwart-data")]
    end
    Browser -- "443 webmail.example.com" --> Caddy
    Browser -- "443 mail.example.com/admin, DAV, autoconfig" --> Caddy
    Caddy -- "http :8080" --> Ihasmail
    Caddy -- "http :8080" --> Stalwart
    Ihasmail -- "JMAP over http://stalwart:8080" --> Stalwart
    Apps -- "25, 465, 993, 995, 4190" --> Stalwart
    Stalwart --- Data
 ```
 - **Caddy** is the only thing on ports 80 and 443. It serves the webmail at the
  webmail host, and Stalwart's web side (admin UI, JMAP for other clients,
  CalDAV, CardDAV, autoconfig, MTA-STS) at the mail host and the
  `autoconfig`, `autodiscover`, `mta-sts` and `ua-auto-config` names.
 - **Stalwart** publishes its mail ports directly, so SMTP and IMAP clients reach
  it with their real addresses.
 - **ihasmail** talks to Stalwart over plain HTTP on the private network, as
  `http://stalwart:8080`. The browser never talks to Stalwart for webmail; it
  talks to ihasmail, and ihasmail makes the JMAP calls.
 - The three containers have **fixed addresses**, because Stalwart is told about
  two of them (see below), and an address Docker picks afresh on every recreate
  couldn't be written into Stalwart's settings.
 ### The deploy, as a sequence
 ```mermaid
 sequenceDiagram
    autonumber
    participant T as ihasmail-oneshot
    participant S as Stalwart
    participant I as ihasmail
    participant C as Caddy
    participant CA as Let's Encrypt
    T->>S: start with a one-time admin (bootstrap mode)
    T->>S: x:Bootstrap/set: hostname, domain, DKIM, logging
    S-->>T: permanent administrator + password
    T->>S: recreate without the one-time admin
    T->>I: start
    T->>C: start
    C->>CA: certificates for its names (TLS-ALPN-01, port 443)
    T->>S: allow ihasmail's address, trust Caddy's X-Forwarded-For
    T->>S: restart to apply
    T->>S: create mailboxes
    T->>I: sign in as the administrator
    I->>S: JMAP session (proves the link)
    T->>S: create ACME account (HTTP-01), domain to automatic certificates
    S->>CA: order certificate
    CA->>C: HTTP-01 challenge on port 80
    C->>S: forwarded /.well-known/acme-challenge/
    S-->>T: certificate issued
    T->>S: read the DNS zone
 ```
 ### Setting up Stalwart without its web wizard
 Stalwart 0.16 keeps its configuration in its data store rather than a file. A
 server started with an empty configuration comes up in **bootstrap mode**: a
 temporary administrator and a setup wizard on port 8080. The wizard is a single
 registry object, `x:Bootstrap`, written with one JMAP call.
 The tool starts Stalwart with `STALWART_RECOVERY_ADMIN` set to a random password
 held only in memory. It's passed through an override file that exists only for
 that step, and through the environment of the `docker compose` process, never
 written into `compose.yaml`. It then sets:
 - `serverHostname` and `defaultDomain` from `--mail-host` and `--domain`;
 - `generateDkimKeys` on, so outgoing mail is signed from the start;
 - the log to **stdout**, because Stalwart's default log directory doesn't exist
  in the container image and isn't a volume;
 - `requestTlsCertificate` **off**, because on 0.16.22 that switch creates no ACME
  account and leaves the domain on manual certificates, so it would look like
  certificates had been handled when they hadn't. The tool sets ACME up itself,
  explicitly.
 Stalwart answers with a permanent administrator, `admin@DOMAIN`, and its
 password. The tool writes that to `credentials.txt` before doing anything else,
 so a failure later can't lose it. Then it brings the full stack up from
 `compose.yaml` alone, which recreates Stalwart without the one-time
 administrator, so no fixed recovery credential outlives the setup.
 The tool refuses to set up a Stalwart that isn't in bootstrap mode, so it can
 never reconfigure a server that already holds someone's mail.
 ### Certificates for two services on one pair of ports
 Caddy needs certificates to serve HTTPS. Stalwart needs its own for IMAP, POP3,
 submission and STARTTLS on SMTP. Both need one for the mail host's name, and
 both would normally want port 443 to prove they control it.
 They're separated by **ACME challenge type**:
 | | Challenge | Port | How |
 | --- | --- | --- | --- |
 | **Caddy** | TLS-ALPN-01 | 443 | Configured with `disable_http_challenge` for Stalwart's names |
 | **Stalwart** | HTTP-01 | 80 | Caddy forwards `/.well-known/acme-challenge/*` for Stalwart's names to Stalwart, untouched. Everything else on port 80 is redirected to HTTPS |
 Neither ever answers the other's challenge, and neither needs the other's key.
 Stalwart's certificate covers the mail host plus `autoconfig`, `autodiscover`,
 `mta-sts` and `ua-auto-config` under the domain, the names its own DNS zone
 points at the mail host, and Caddy fronts all five so every challenge reaches
 it.
 Stalwart starts its order as soon as the ACME account exists, and renews by
 itself from then on. If an order fails, usually because DNS isn't in place yet,
 Stalwart doesn't try again by itself. `certs` starts a new order.
 ### Stalwart's automatic IP bans, behind a proxy
 Stalwart bans an IP address that behaves like an attacker. For example, 30
 requests in a day for scanner paths such as `/wp-admin.php` get an address
 banned. That's good protection, but it counts by the address a connection comes
 from, and two kinds of traffic reach Stalwart from a single shared address:
 1. **Everything through Caddy** arrives from Caddy's address. Unchanged, one bot
   scanning `https://mail.example.com` bans Caddy. That was reproduced on
   0.16.22: after a scan, autoconfig answered `403` to everyone, and so would
   calendar sync and certificate renewal.
   **Fix:** Stalwart is told to take the client's address from the
   `X-Forwarded-For` header Caddy sets. Re-tested after the change, the scanner
   was banned and every other client still got through. This is safe only
   because nothing untrusted can reach Stalwart's HTTP port: it's published on
   loopback alone, and on the private network its only peers are Caddy, which
   sets the header itself, and ihasmail, which sends none.
 2. **Every webmail request** arrives from ihasmail's address: every sign-in, and
   every push stream opened and dropped as people open and close tabs. A ban on
   that one address would be a ban on everyone's webmail.
   **Fix:** ihasmail's fixed address is added to Stalwart's allowed addresses.
   ihasmail limits sign-in attempts per real client itself, so repeated wrong
   passwords are still slowed down, just not by banning the webmail.
 On 0.16.22 **neither setting takes effect until Stalwart restarts**. Applied to
 a running server, a scan straight afterwards still banned Caddy. So the tool
 restarts Stalwart after making both changes, before anything else depends on
 them.
 ### The webmail's side
 ihasmail is deployed the way its own documentation recommends for this
 situation:
 - **Immutable**: read-only root filesystem, no volume, sessions in memory. A
  restart signs everyone out and loses nothing else, because everything durable,
  each user's settings included, lives in Stalwart.
 - **Private route to Stalwart** (`STALWART_URL=http://stalwart:8080`): the
  credentials that travel on that leg never leave the host.
 - **Push by subscription** (`PUSH_URL=https://webmail.example.com`): Stalwart
  posts mailbox changes to ihasmail instead of holding a connection open for
  every browser tab. If Stalwart can't reach that URL, every tab uses the
  ordinary relay instead and nothing breaks. `/api/health` shows which is in use.
 - **Behind a trusted proxy** (`TRUST_PROXY=1`): ihasmail believes Caddy's
  forwarded headers, because Caddy is on a private network range, so it sets
  secure cookies and attributes sign-in attempts to the real client.
 ## Security model
 **What's reachable from outside** (mail host shape):
 | Port | Service |
 | --- | --- |
 | 25 | SMTP, Stalwart (receiving mail; STARTTLS) |
 | 80 | Caddy: redirects to HTTPS, and ACME HTTP-01 challenges for Stalwart |
 | 443 (TCP, UDP) | Caddy: the webmail, and Stalwart's web side |
 | 465 | SMTP submission with TLS, Stalwart |
 | 993 | IMAP with TLS, Stalwart |
 | 995 | POP3 with TLS, Stalwart |
 | 4190 | ManageSieve, Stalwart |
 Stalwart's plain-HTTP port (8080) and ihasmail's port are published on
 `127.0.0.1` only. In `--local` mode, those two loopback ports are all that's
 published.
 **Secrets:**
 - `APP_SECRET` is 48 random bytes, in `.env` (0600). It seals ihasmail's session
  cookies.
 - Generated passwords come from the operating system's cryptographic random
  source, letters and digits only, in `credentials.txt` (0600).
 - The one-time bootstrap password is never written to disk, and Stalwart is
  recreated without it once setup completes.
 **Trust decisions the deployment makes**, each explained above:
 - Stalwart believes `X-Forwarded-For` on its HTTP port, which only Caddy and
  ihasmail can reach.
 - ihasmail's address is exempt from Stalwart's automatic bans.
 - ihasmail believes forwarded headers from private-range peers, which here is
  Caddy.
 **What the tool doesn't do:** configure a host firewall, harden the Docker
 daemon, set up backups or monitoring, or turn on encryption at rest for
 mailboxes. Encryption at rest can't be turned off again once on, which is not a
 decision for a deploy tool to make.
 ## Troubleshooting
 **`port 443 is already in use on this host`**
 Something else, often an existing web server, holds a port the mail host needs.
 Free it, or use a separate host. With `--local`, move `--webmail-bind` or
 `--stalwart-bind` instead.
 **`project ihasmail-example-com already exists in Docker`**
 An earlier run left containers or volumes behind. If it's a failed attempt you
 want to discard: `ihasmail-oneshot destroy --dir ihasmail-example-com --yes`,
 then deploy again. To keep it and deploy another, pass a different `--project`
 and `--dir`.
 **`... already has files in it; give --dir a new or empty directory`**
 The tool never writes over an existing deployment. Choose another directory, or
 destroy the old deployment first.
 **A deploy stopped partway.**
 The error says which step failed and shows the last lines of the relevant
 container's log. The stack is left as it was, for you to inspect. To start
 again from nothing: `ihasmail-oneshot destroy --dir DIR --yes`, fix the cause,
 and deploy again.
 **`Pool overlaps with other one on this address space`**
 The default private network `172.31.253.0/24` collides with a Docker network you
 already have. Destroy the partial deployment and deploy again with, for
 example, `--subnet 172.31.200.0/24`.
 **`Stalwart has no certificate yet`**
 Usually DNS: the mail host or one of the `autoconfig`, `autodiscover`,
 `mta-sts`, `ua-auto-config` names doesn't resolve to this host yet, or port 80
 isn't reachable from the internet. Check with `dig +short mail.example.com` from
 somewhere else, fix it, then run `ihasmail-oneshot certs --dir DIR`. Stalwart's
 reasons are in its log:
 ```bash
 docker compose logs stalwart | grep -i acme
 ```
 **`Caddy has no certificate for webmail.example.com yet`**
 Same causes. Caddy retries by itself with increasing delays. See
 `docker compose logs caddy | grep -i error`.
 **Mail arrives but sent mail never does.**
 Outbound port 25 is almost always the cause. Test from the host with
 `nc -vz gmail-smtp-in.l.google.com 25`. Stalwart's log names each delivery
 failure (`docker compose logs stalwart`). Also check that reverse DNS for the host's address
 names the mail host.
 **Sent mail lands in spam.**
 Check that the SPF, DKIM and DMARC records from `dns-records.zone` are
 published exactly, and that reverse DNS is set. New addresses need a little time
 to build a sending reputation.
 **The webmail signs everyone out.**
 Expected after `docker compose restart ihasmail`, any ihasmail upgrade, or a
 reboot: sessions live in memory.
 **`/api/health` shows push accounts `pending` that never become `verified`.**
 Stalwart can't reach `https://webmail.example.com` from inside its container.
 Mail still updates in real time over the relay. The usual causes are a host
 firewall that drops traffic from Docker's bridge to the host's own public
 address, or the webmail name not resolving from the host.
 ## Known limits
 - **One mail domain per deploy.** More can be added afterwards in Stalwart or in
  ihasmail's Administration. Their certificates and DNS records are then yours
  to arrange.
 - **Linux Docker hosts only.** The tool drives the local Docker daemon and
  publishes ports on the host it runs on.
 - **IPv4 on the private network.** Ports are also published on IPv6 wherever
  Docker does so on the host.
 - **Fresh deployments only.** It doesn't import an existing Stalwart, or adopt a
  deployment it didn't write.
 - **Stalwart doesn't retry a failed certificate order** by itself, on a
  transient CA error either. `certs` starts a new one.
 - **Push by subscription isn't covered by the end-to-end test**, whose lab has no
  public DNS for Stalwart to resolve the webmail's name through. Its fallback,
  the relay, is what the test exercises.
 - Validated against **Stalwart 0.16.22**. Other Stalwart versions may change the
  registry objects the setup uses.
 ## Development and testing
 ```bash
 go vet ./...
 go test ./...          # unit tests: validation, rendered files, the Stalwart client
 e2e/public.sh          # the whole mail-host deploy, for real, on this machine
 ```
 `e2e/public.sh` runs a complete mail-host deployment with no internet involved.
 [Pebble](https://github.com/letsencrypt/pebble), the ACME test server, stands in
 for Let's Encrypt, and a DNS stub answers every name with the host's own address.
 Caddy and Stalwart both obtain real certificates from it, through the real
 ports and the real `Caddyfile`. It then checks, among other things, that:
 - the deploy completes, links ihasmail and Stalwart, and reports both kinds of
  certificate;
 - `credentials.txt` and `.env` are private, and the bootstrap credential is gone
  from the Stalwart container;
 - Stalwart's IMAPS (993) and submission (465) ports present certificates that
  verify for the mail host;
 - the webmail, Stalwart's JMAP and autoconfig are served over verified HTTPS, and
  a user signs in through the webmail;
 - `dns-records.zone` holds the MX and DKIM records, and `certs` recognises an
  existing certificate;
 - a scanner probing through Caddy is banned by its own address, not Caddy's, and
  other clients still get through;
 - ihasmail is exempt from bans, and a user still signs in after failed attempts.
 It publishes ports 25, 80, 443, 465, 993, 995 and 4190 on the machine while it
 runs, and removes everything it created when it ends, pass or fail. `KEEP=1
 e2e/public.sh` leaves the stack up to inspect.
 The code:
 | Package | Does |
 | --- | --- |
 | `cmd/ihasmail-oneshot` | Commands, flags, confirmation, the summary |
 | `internal/config` | Validates the command line into a plan: names, addresses, images |
 | `internal/render` | Renders `compose.yaml` and the `Caddyfile`; writes files without ever overwriting |
 | `internal/docker` | Drives the `docker` and `docker compose` CLIs |
 | `internal/stalwart` | The JMAP client and every Stalwart registry call: bootstrap, ACME, bans, mailboxes, DNS zone |
 | `internal/webmail` | ihasmail's health check and the sign-in that proves the link |
 | `internal/deploy` | Preflight, the deploy sequence, `certs`, `destroy` |
 ## Versions and releases
 Releases are tagged by date, like ihasmail's: `v2026.9.13`, with `.1`, `.2`
 added for another release the same day. Each release pins the Stalwart,
 ihasmail and Caddy images it was tested with as its defaults. A newer release
 of the tool generally means newer tested versions.
 Binaries for `linux/amd64` and `linux/arm64` and a `SHA256SUMS` file are
 attached to every [release](https://github.com/Coffey-Labs/ihasmail-oneshot/releases).
 ## Security
@@ -872,9 +99,6 @@ public issue.
 AGPL-3.0-or-later, the same as ihasmail. See [LICENSE](LICENSE).
 Running the tool to deploy your own mail host places no obligations on you. The
-licence matters if you modify the tool and offer it to others, including as a
+license matters if you modify the tool and offer it to others, including as a
 hosted service that deploys on their behalf: then your modified source must be
 available to them.
 `v2026.9.13`, the first release, was published under GPL-3.0-or-later and
 stays under it for anyone who has it. Every later release is AGPL-3.0-or-later.
@@ -101,7 +101,7 @@ Flags:
 	fs.StringVar(&o.Dir, "dir", "", "deployment directory to write, new or empty (default ./PROJECT)")
 	fs.StringVar(&o.Project, "project", "", "compose project name (default ihasmail-DOMAIN, dots as dashes)")
 	fs.StringVar(&o.StalwartImage, "stalwart-image", config.DefaultStalwartImage, "Stalwart image")
-	fs.StringVar(&o.IhasmailImage, "ihasmail-image", config.DefaultIhasmailImage, "ihasmail image")
+	fs.StringVar(&o.IhasmailImage, "ihasmail-image", config.NewestIhasmail, "ihasmail image; the default is the newest release, written into compose.yaml as its dated tag")
 	fs.StringVar(&o.CaddyImage, "caddy-image", config.DefaultCaddyImage, "Caddy image")
 	fs.StringVar(&o.WebmailBind, "webmail-bind", "127.0.0.1:8080", "host address for ihasmail's own port")
 	fs.StringVar(&o.StalwartBind, "stalwart-bind", "127.0.0.1:8081", "host address for Stalwart's plain-HTTP port (admin UI)")
@@ -132,6 +132,10 @@ Flags:
 	for _, w := range warnings {
 		log.Warn("%s", w)
 	}
 	// Before the question, so the answer is about the exact ihasmail release.
 	if plan, err = deploy.ResolveIhasmail(ctx, plan, log); err != nil {
 		return err
 	}
 	if !yes {
 		if err := confirm("deploy this?"); err != nil {
 			return err
@@ -142,7 +146,7 @@ Flags:
 	if err != nil {
 		return err
 	}
-	summarise(plan, res, log)
+	summarize(plan, res, log)
 	return nil
 }
@@ -162,7 +166,11 @@ func describe(p config.Plan, log deploy.Log) {
 			log.Info("ACME      Let's Encrypt, contact %s", p.Email)
 		}
 	}
-	log.Info("images    %s, %s", p.StalwartImage, p.IhasmailImage)
+	if p.FollowsNewestIhasmail() {
 		log.Info("images    %s, ihasmail's newest release", p.StalwartImage)
 	} else {
 		log.Info("images    %s, %s", p.StalwartImage, p.IhasmailImage)
 	}
 	if !p.Local {
 		log.Info("          %s", p.CaddyImage)
 	}
@@ -171,7 +179,7 @@ func describe(p config.Plan, log deploy.Log) {
 	}
 }
-func summarise(p config.Plan, r *deploy.Result, log deploy.Log) {
+func summarize(p config.Plan, r *deploy.Result, log deploy.Log) {
 	log.Step("done")
 	if p.Local {
 		log.Info("webmail      http://%s", p.WebmailBind)
@@ -0,0 +1,275 @@
 # How ihasmail-oneshot works
 The design of the deployment this tool writes, and the reasons behind each
 choice: what runs where, the order the deploy happens in, how Stalwart is set up
 without its web wizard, how two services share certificates on one pair of
 ports, and how Stalwart's automatic bans are kept from locking everyone out. For
 installing and using the tool, start with the [README](../README.md) or the
 [guide on docs.ihasmail.org](https://docs.ihasmail.org/install/oneshot/).
 ## Why it exists
 Stalwart and ihasmail each install easily on their own. Making them a working,
 safe mail host *together* takes a series of decisions that are easy to get
 wrong, and most of the mistakes don't show until later:
 - **Stalwart 0.16 has no configuration file to template.** Its settings live in
  its data store, and a new server starts in a bootstrap mode that expects a
  person at a web wizard. Automating that means driving its registry API, and
  making sure the bootstrap credential doesn't quietly outlive the setup.
 - **Both services want the same ports for certificates.** The webmail needs
  HTTPS on 443. Stalwart needs its own certificate for IMAP and SMTP, and its
  default way of getting one also wants port 443. Two ACME clients fighting over
  one port fail in confusing, intermittent ways.
 - **Stalwart's automatic IP bans don't expect a proxy in front of it.** Stalwart
  bans addresses that probe for things like WordPress admin pages. Behind a
  reverse proxy, every client arrives from the proxy's address, so one bot
  scanning your mail host gets *the proxy* banned. Autoconfig, calendar sync and
  certificate renewals all stop working, for everyone, with nothing obviously
  wrong. This was reproduced on Stalwart 0.16.22 while building this tool, as
  was a second trap: the setting that fixes it only takes effect after a
  restart.
 - **The route from webmail to mail server matters.** Sent over the private
  Docker network, it never leaves the host and costs about a third of the
  memory per signed-in browser tab that going back out over public HTTPS does.
 - **Mail doesn't flow until DNS is right**, and the list of records (MX, SPF,
  DKIM, DMARC, SRV, MTA-STS, autoconfig) is long.
 This tool makes each of those decisions once, the same way every time, and
 checks the result before telling you it is done. Each one is explained below.
 ## What it deploys
 A single, statically linked binary for Linux (amd64 and arm64). You run it on
 the Docker host that should become the mail server. It needs Docker and the
 compose plugin, and nothing else: no configuration file to write first, no
 language runtime, no other scripts.
 It deploys three containers:
 | Container | Image | Role |
 | --- | --- | --- |
 | **Stalwart** | `stalwartlabs/stalwart` | The mail server: SMTP, IMAP, POP3, JMAP, CalDAV, CardDAV, spam filtering, DKIM. It holds all the mail and all the accounts |
 | **ihasmail** | `ghcr.io/coffey-labs/ihasmail` | The webmail: mail, calendars, contacts, files and filters in the browser, talking to Stalwart over JMAP. It holds nothing but sessions |
 | **Caddy** | `caddy` | The HTTPS front: certificates and TLS for the webmail and for Stalwart's web side |
 It has two shapes:
 | | `deploy --domain example.com` | `deploy --local` |
 | --- | --- | --- |
 | **For** | A real mail host on the internet | Trying ihasmail against a real Stalwart on your own machine |
 | **Containers** | Stalwart, ihasmail, Caddy | Stalwart, ihasmail |
 | **Publishes** | 25, 80, 443, 465, 993, 995, 4190 on every interface | Nothing but two loopback ports |
 | **Certificates** | Let's Encrypt, for both Caddy and Stalwart | None |
 | **Webmail** | `https://webmail.example.com` | `http://127.0.0.1:8080` |
 | **Stalwart admin** | `https://mail.example.com/admin` | `http://127.0.0.1:8081/admin` |
 ## What a deploy does, in order
 In order, in one run:
 1. **Checks the host** before changing anything: Docker and compose are there
   and usable, every port it needs is free, the deployment directory is new or
   empty, no compose project of the same name exists, and the hostnames resolve.
   It reports every problem at once, not the first.
 2. **Finds ihasmail's newest release, shows you the plan and asks** before
   going ahead, so what you agree to is an exact ihasmail version. `--yes` skips the
   question, and is required when there is no terminal to ask on.
 3. **Writes the deployment directory**: `compose.yaml`, the `Caddyfile`, and
   `.env` holding a freshly generated `APP_SECRET`.
 4. **Pulls the images**: Stalwart and Caddy at the versions this release of the
   tool was tested with, and the ihasmail release it found, written into
   `compose.yaml` by its dated tag so nothing moves it later.
 5. **Starts Stalwart in bootstrap mode** with a one-time administrator whose
   password exists only in the tool's memory.
 6. **Completes Stalwart's setup** through its API, the same setup its web UI
   would otherwise walk you through: hostname, mail domain, DKIM keys, logging.
   Stalwart creates the permanent administrator and returns its password, which
   is written to `credentials.txt` straight away.
 7. **Starts the whole stack.** Stalwart is recreated *without* the one-time
   administrator.
 8. **Links ihasmail and Stalwart**: exempts ihasmail from Stalwart's automatic
   IP bans, tells Stalwart to trust the client addresses Caddy forwards, and
   restarts Stalwart so both apply.
 9. **Creates the mailboxes** you asked for with `--user`, each with a generated
   password.
 10. **Proves the link** by signing in *through ihasmail* as the administrator.
    That only succeeds if ihasmail can reach Stalwart and Stalwart accepts the
    credentials.
 11. **Requests certificates** (mail host only): sets up an ACME account in
    Stalwart for its IMAP and SMTP certificate, and waits for that and for
    Caddy's certificates to arrive.
 12. **Writes `dns-records.zone`**, every DNS record Stalwart wants published,
    and prints a summary of URLs, credentials and anything still to do.
 ## The shape of the deployment
 ```mermaid
 flowchart LR
    Browser(["Browser"])
    Apps(["Mail apps, other mail servers"])
    subgraph Host["Docker host"]
        subgraph Net["private network 172.31.253.0/24"]
            Caddy["Caddy<br/>.10"]
            Ihasmail["ihasmail<br/>.11<br/>read-only, no volume"]
            Stalwart["Stalwart<br/>.12"]
        end
        Data[("stalwart-data")]
    end
    Browser -- "443 webmail.example.com" --> Caddy
    Browser -- "443 mail.example.com/admin, DAV, autoconfig" --> Caddy
    Caddy -- "http :8080" --> Ihasmail
    Caddy -- "http :8080" --> Stalwart
    Ihasmail -- "JMAP over http://stalwart:8080" --> Stalwart
    Apps -- "25, 465, 993, 995, 4190" --> Stalwart
    Stalwart --- Data
 ```
 - **Caddy** is the only thing on ports 80 and 443. It serves the webmail at the
  webmail host, and Stalwart's web side (admin UI, JMAP for other clients,
  CalDAV, CardDAV, autoconfig, MTA-STS) at the mail host and the
  `autoconfig`, `autodiscover`, `mta-sts` and `ua-auto-config` names.
 - **Stalwart** publishes its mail ports directly, so SMTP and IMAP clients reach
  it with their real addresses.
 - **ihasmail** talks to Stalwart over plain HTTP on the private network, as
  `http://stalwart:8080`. The browser never talks to Stalwart for webmail; it
  talks to ihasmail, and ihasmail makes the JMAP calls.
 - The three containers have **fixed addresses**, because Stalwart is told about
  two of them (see below), and an address Docker picks afresh on every recreate
  couldn't be written into Stalwart's settings.
 ## The deploy, as a sequence
 ```mermaid
 sequenceDiagram
    autonumber
    participant T as ihasmail-oneshot
    participant S as Stalwart
    participant I as ihasmail
    participant C as Caddy
    participant CA as Let's Encrypt
    T->>S: start with a one-time admin (bootstrap mode)
    T->>S: x:Bootstrap/set: hostname, domain, DKIM, logging
    S-->>T: permanent administrator + password
    T->>S: recreate without the one-time admin
    T->>I: start
    T->>C: start
    C->>CA: certificates for its names (TLS-ALPN-01, port 443)
    T->>S: allow ihasmail's address, trust Caddy's X-Forwarded-For
    T->>S: restart to apply
    T->>S: create mailboxes
    T->>I: sign in as the administrator
    I->>S: JMAP session (proves the link)
    T->>S: create ACME account (HTTP-01), domain to automatic certificates
    S->>CA: order certificate
    CA->>C: HTTP-01 challenge on port 80
    C->>S: forwarded /.well-known/acme-challenge/
    S-->>T: certificate issued
    T->>S: read the DNS zone
 ```
 ## Setting up Stalwart without its web wizard
 Stalwart 0.16 keeps its configuration in its data store rather than a file. A
 server started with an empty configuration comes up in **bootstrap mode**: a
 temporary administrator and a setup wizard on port 8080. The wizard is a single
 registry object, `x:Bootstrap`, written with one JMAP call.
 The tool starts Stalwart with `STALWART_RECOVERY_ADMIN` set to a random password
 held only in memory. It's passed through an override file that exists only for
 that step, and through the environment of the `docker compose` process, never
 written into `compose.yaml`. It then sets:
 - `serverHostname` and `defaultDomain` from `--mail-host` and `--domain`;
 - `generateDkimKeys` on, so outgoing mail is signed from the start;
 - the log to **stdout**, because Stalwart's default log directory doesn't exist
  in the container image and isn't a volume;
 - `requestTlsCertificate` **off**, because on 0.16.22 that switch creates no ACME
  account and leaves the domain on manual certificates, so it would look like
  certificates had been handled when they hadn't. The tool sets ACME up itself,
  explicitly.
 Stalwart answers with a permanent administrator, `admin@DOMAIN`, and its
 password. The tool writes that to `credentials.txt` before doing anything else,
 so a failure later can't lose it. Then it brings the full stack up from
 `compose.yaml` alone, which recreates Stalwart without the one-time
 administrator, so no fixed recovery credential outlives the setup.
 The tool refuses to set up a Stalwart that isn't in bootstrap mode, so it can
 never reconfigure a server that already holds someone's mail.
 ## Certificates for two services on one pair of ports
 Caddy needs certificates to serve HTTPS. Stalwart needs its own for IMAP, POP3,
 submission and STARTTLS on SMTP. Both need one for the mail host's name, and
 both would normally want port 443 to prove they control it.
 They're separated by **ACME challenge type**:
 | | Challenge | Port | How |
 | --- | --- | --- | --- |
 | **Caddy** | TLS-ALPN-01 | 443 | Configured with `disable_http_challenge` for Stalwart's names |
 | **Stalwart** | HTTP-01 | 80 | Caddy forwards `/.well-known/acme-challenge/*` for Stalwart's names to Stalwart, untouched. Everything else on port 80 is redirected to HTTPS |
 Neither ever answers the other's challenge, and neither needs the other's key.
 Stalwart's certificate covers the mail host plus `autoconfig`, `autodiscover`,
 `mta-sts` and `ua-auto-config` under the domain, the names its own DNS zone
 points at the mail host, and Caddy fronts all five so every challenge reaches
 it.
 Stalwart starts its order as soon as the ACME account exists, and renews by
 itself from then on. If an order fails, usually because DNS isn't in place yet,
 Stalwart doesn't try again by itself. `certs` starts a new order.
 ## Stalwart's automatic IP bans, behind a proxy
 Stalwart bans an IP address that behaves like an attacker. For example, 30
 requests in a day for scanner paths such as `/wp-admin.php` get an address
 banned. That's good protection, but it counts by the address a connection comes
 from, and two kinds of traffic reach Stalwart from a single shared address:
 1. **Everything through Caddy** arrives from Caddy's address. Unchanged, one bot
   scanning `https://mail.example.com` bans Caddy. That was reproduced on
   0.16.22: after a scan, autoconfig answered `403` to everyone, and so would
   calendar sync and certificate renewal.
   **Fix:** Stalwart is told to take the client's address from the
   `X-Forwarded-For` header Caddy sets. Re-tested after the change, the scanner
   was banned and every other client still got through. This is safe only
   because nothing untrusted can reach Stalwart's HTTP port: it's published on
   loopback alone, and on the private network its only peers are Caddy, which
   sets the header itself, and ihasmail, which sends none.
 2. **Every webmail request** arrives from ihasmail's address: every sign-in, and
   every push stream opened and dropped as people open and close tabs. A ban on
   that one address would be a ban on everyone's webmail.
   **Fix:** ihasmail's fixed address is added to Stalwart's allowed addresses.
   ihasmail limits sign-in attempts per real client itself, so repeated wrong
   passwords are still slowed down, just not by banning the webmail.
 On 0.16.22 **neither setting takes effect until Stalwart restarts**. Applied to
 a running server, a scan straight afterwards still banned Caddy. So the tool
 restarts Stalwart after making both changes, before anything else depends on
 them.
 ## The webmail's side
 ihasmail is deployed the way its own documentation recommends for this
 situation:
 - **Immutable**: read-only root filesystem, no volume, sessions in memory. A
  restart signs everyone out and loses nothing else, because everything durable,
  each user's settings included, lives in Stalwart.
 - **Private route to Stalwart** (`STALWART_URL=http://stalwart:8080`): the
  credentials that travel on that leg never leave the host.
 - **Push by subscription** (`PUSH_URL=https://webmail.example.com`): Stalwart
  posts mailbox changes to ihasmail instead of holding a connection open for
  every browser tab. If Stalwart can't reach that URL, every tab uses the
  ordinary relay instead and nothing breaks. `/api/health` shows which is in use.
 - **Behind a trusted proxy** (`TRUST_PROXY=1`): ihasmail believes Caddy's
  forwarded headers, because Caddy is on a private network range, so it sets
  secure cookies and attributes sign-in attempts to the real client.
@@ -0,0 +1,204 @@
 # Reference
 Everything an operator looks up: requirements, installing, each command and
 flag, what the deployment directory holds, and running, upgrading, backing up
 and removing a deployment. For a walk-through, see the
 [guide on docs.ihasmail.org](https://docs.ihasmail.org/install/oneshot/). Back
 to the [README](../README.md).
 ## Requirements
 **On the host:**
 - Linux, amd64 or arm64.
 - Docker Engine with the compose plugin (`docker compose version` works), run
  as a user allowed to use Docker.
 **For a mail host, additionally:**
 - A domain whose DNS you control.
 - **Ports 25, 80, 443, 465, 993, 995 and 4190** free on the host and open in any
  firewall or cloud security group in front of it.
 - **Outbound port 25.** Many cloud and VPS providers block it by default and
  unblock it on request. Without it you can receive mail but not send it.
 - **Reverse DNS**: a PTR record for the host's IP address naming the mail host
  (`mail.example.com`). This is set at your hosting provider, not in your DNS
  zone. Many receiving servers reject mail from an address without one.
 - A static public IP address.
 **Resources:** ihasmail on its own needs 256 MiB of RAM at minimum. Stalwart's
 needs grow with the mail it stores and the number of people using it; size the
 host for Stalwart.
 ## Installing
 ### From a release
 Download the archive for your architecture and its checksums, verify, and
 unpack:
 ```bash
 ARCH=amd64   # or arm64
 curl -fsSLO https://github.com/Coffey-Labs/ihasmail-oneshot/releases/latest/download/ihasmail-oneshot-linux-$ARCH.tar.gz
 curl -fsSLO https://github.com/Coffey-Labs/ihasmail-oneshot/releases/latest/download/SHA256SUMS
 sha256sum --ignore-missing -c SHA256SUMS
 tar -xzf ihasmail-oneshot-linux-$ARCH.tar.gz
 sudo install -m 0755 ihasmail-oneshot /usr/local/bin/
 ihasmail-oneshot version
 ```
 Every release is built by the [release workflow](../.github/workflows/release.yml)
 from a tagged commit on `main`, after the tests and a known-vulnerabilities
 check pass. The archives are reproducible: `scripts/build-release.sh` builds the
 same bytes from the same commit.
 ### From source
 With Go 1.26.8 or newer:
 ```bash
 git clone https://github.com/Coffey-Labs/ihasmail-oneshot.git
 cd ihasmail-oneshot
 go build -o ihasmail-oneshot ./cmd/ihasmail-oneshot
 ```
 ## Commands
 ### `deploy`
 ```text
 ihasmail-oneshot deploy --domain example.com [flags]
 ihasmail-oneshot deploy --local [flags]
 ```
 | Flag | Default | Meaning |
 | --- | --- | --- |
 | `--domain` | required, or `example.test` with `--local` | The mail domain this server receives for |
 | `--local` | off | The loopback-only shape: no mail ports, no Caddy, no certificates |
 | `--mail-host` | `mail.DOMAIN` | Stalwart's hostname. Must be one label under the domain, e.g. `mx.example.com` |
 | `--webmail-host` | `webmail.DOMAIN` | The webmail's hostname. Any name except Stalwart's |
 | `--email` | `postmaster@DOMAIN` | ACME contact address for both Caddy and Stalwart |
 | `--user NAME` | none | Create a mailbox `NAME@DOMAIN` with a generated password. Repeat for more |
 | `--dir` | `./PROJECT` | Deployment directory to write. Must be new or empty |
 | `--project` | `ihasmail-DOMAIN` (dots as dashes) | Compose project name, which prefixes containers, network and volumes |
 | `--stalwart-image` | `stalwartlabs/stalwart:v0.16.22` | Stalwart image |
 | `--ihasmail-image` | the newest release | ihasmail image. By default the tool looks up ihasmail's newest release and writes it into `compose.yaml` by its dated tag; name an image to use a particular one |
 | `--caddy-image` | `caddy:2.11.4` | Caddy image |
 | `--webmail-bind` | `127.0.0.1:8080` | Host address for ihasmail's own port, for reaching it without Caddy |
 | `--stalwart-bind` | `127.0.0.1:8081` | Host address for Stalwart's plain-HTTP port. The tool configures Stalwart through it |
 | `--subnet` | `172.31.253.0/24` | The stack's private network. Change it if it overlaps a network you already have |
 | `--acme-directory` | Let's Encrypt | ACME directory URL of a private CA, for both Caddy and Stalwart |
 | `--acme-ca-root` | none | PEM file of the root the private CA's HTTPS endpoint is signed by |
 | `--yes` | off | Don't ask for confirmation. Required without a terminal |
 Stalwart's and Caddy's defaults are the versions tested together for this release. ihasmail's default is its newest release, looked up when the tool runs and written into `compose.yaml` by its dated tag.
 ### `certs`
 ```text
 ihasmail-oneshot certs --dir DIR
 ```
 Starts a new certificate order in Stalwart for the deployment in `DIR` and waits
 for it, typically after fixing DNS. If Stalwart already holds a valid
 certificate for the mail host, it reports that and does nothing.
 ### `destroy`
 ```text
 ihasmail-oneshot destroy --dir DIR [--yes]
 ```
 Removes the deployment in `DIR` with all of its data. See
 [Removing a deployment](#removing-a-deployment).
 ### `version`
 Prints the tool's version.
 ## The deployment directory
 The deploy writes a directory named after the domain, `./ihasmail-example-com`
 by default (`--dir` to choose):
 | File | Mode | What it is |
 | --- | --- | --- |
 | `compose.yaml` | 0644 | The whole deployment: three services, a private network, named volumes. Commented |
 | `Caddyfile` | 0644 | Caddy's configuration: the webmail, Stalwart's web names, and port 80's challenge forwarding |
 | `.env` | 0600 | `APP_SECRET`, which seals ihasmail's session cookies. Read by compose |
 | `credentials.txt` | 0600 | The administrator's and each mailbox's generated password, plus what `certs` needs to reach Stalwart |
 | `dns-records.zone` | 0644 | The DNS records to publish |
 | `acme-ca-root.pem`, `ca-bundle.crt` | 0644 | Only with `--acme-ca-root`: the private CA, and the system roots with it added |
 Data lives in Docker named volumes, prefixed with the project name:
 | Volume | Holds |
 | --- | --- |
 | `…_stalwart-data` | All mail, accounts, calendars, contacts, files, settings, DKIM keys, Stalwart's certificates |
 | `…_stalwart-etc` | Stalwart's store location file |
 | `…_caddy-data` | Caddy's certificates and ACME account |
 | `…_caddy-config` | Caddy's autosaved configuration |
 ihasmail has no volume. It runs read-only and keeps sessions in memory.
 ## Running a deployment
 The directory is a standard compose project. From inside it:
 ```bash
 docker compose ps                      # what's running
 docker compose logs -f stalwart        # follow Stalwart's log
 docker compose logs caddy | grep -i error
 docker compose restart ihasmail        # signs everyone out of the webmail; nothing else is lost
 docker compose down                    # stop everything (data stays in the volumes)
 docker compose up -d                   # start it again
 ```
 The containers restart by themselves after a crash or a reboot
 (`restart: unless-stopped`).
 ### Upgrading
 Nothing upgrades on its own: every image in `compose.yaml` is a fixed version,
 ihasmail included. To upgrade, change the image tag there and apply it:
 ```bash
 docker compose pull && docker compose up -d
 ```
 - **ihasmail** is safe to move to any newer release that supports your
  Stalwart version. Its release notes say which. The newest is on
  [ihasmail's releases](https://github.com/Coffey-Labs/ihasmail/releases); its
  image tag is the version with `+` written as `-`, e.g. `2026.9.13-pr344`.
 - **Stalwart**: read its upgrade notes before changing versions. Check that the
  ihasmail version you run supports the new Stalwart release first, since
  ihasmail validates against one Stalwart release at a time. Back up
  `stalwart-data` first.
 - **Caddy** minor releases are routine.
 ### Backing up
 Everything that can't be recreated is in the `stalwart-data` volume. For a
 consistent copy, stop Stalwart briefly:
 ```bash
 docker compose stop stalwart
 docker run --rm -v ihasmail-example-com_stalwart-data:/data -v "$PWD":/backup alpine \
  tar -czf /backup/stalwart-data-$(date +%F).tar.gz -C /data .
 docker compose start stalwart
 ```
 Keep `caddy-data` too, or certificates are requested again on a rebuild. That's
 harmless unless it happens often enough to meet Let's Encrypt's rate limits. Keep
 the deployment directory itself, which is small and contains your secrets.
 ### Removing a deployment
 ```bash
 ihasmail-oneshot destroy --dir ihasmail-example-com
 ```
 This removes the containers, the network, **the volumes with every message and
 account**, and the files the tool wrote. It asks first unless given `--yes`.
 The directory goes too, unless it holds files you added yourself; those are left
 in place, and so is the directory around them.
@@ -0,0 +1,44 @@
 # Security model
 What a deployment exposes to the internet, where its secrets live, and the
 trust decisions it makes on your behalf. The reasoning behind those decisions is
 in [How it works](how-it-works.md). To report a vulnerability, see
 [SECURITY.md](../SECURITY.md). Back to the [README](../README.md).
 **What's reachable from outside** (mail host shape):
 | Port | Service |
 | --- | --- |
 | 25 | SMTP, Stalwart (receiving mail; STARTTLS) |
 | 80 | Caddy: redirects to HTTPS, and ACME HTTP-01 challenges for Stalwart |
 | 443 (TCP, UDP) | Caddy: the webmail, and Stalwart's web side |
 | 465 | SMTP submission with TLS, Stalwart |
 | 993 | IMAP with TLS, Stalwart |
 | 995 | POP3 with TLS, Stalwart |
 | 4190 | ManageSieve, Stalwart |
 Stalwart's plain-HTTP port (8080) and ihasmail's port are published on
 `127.0.0.1` only. In `--local` mode, those two loopback ports are all that's
 published.
 **Secrets:**
 - `APP_SECRET` is 48 random bytes, in `.env` (0600). It seals ihasmail's session
  cookies.
 - Generated passwords come from the operating system's cryptographic random
  source, letters and digits only, in `credentials.txt` (0600).
 - The one-time bootstrap password is never written to disk, and Stalwart is
  recreated without it once setup completes.
 **Trust decisions the deployment makes**, each explained in [How it works](how-it-works.md):
 - Stalwart believes `X-Forwarded-For` on its HTTP port, which only Caddy and
  ihasmail can reach.
 - ihasmail's address is exempt from Stalwart's automatic bans.
 - ihasmail believes forwarded headers from private-range peers, which here is
  Caddy.
 **What the tool doesn't do:** configure a host firewall, harden the Docker
 daemon, set up backups or monitoring, or turn on encryption at rest for
 mailboxes. Encryption at rest can't be turned off again once on, which is not a
 decision for a deploy tool to make.
@@ -0,0 +1,89 @@
 # Troubleshooting and known limits
 Problems by the message or symptom you see, what causes each, and what to do;
 then the limits of what the tool does. Back to the [README](../README.md), or
 see the [guide on docs.ihasmail.org](https://docs.ihasmail.org/install/oneshot/).
 ## Problems
 **`port 443 is already in use on this host`**
 Something else, often an existing web server, holds a port the mail host needs.
 Free it, or use a separate host. With `--local`, move `--webmail-bind` or
 `--stalwart-bind` instead.
 **`project ihasmail-example-com already exists in Docker`**
 An earlier run left containers or volumes behind. If it's a failed attempt you
 want to discard: `ihasmail-oneshot destroy --dir ihasmail-example-com --yes`,
 then deploy again. To keep it and deploy another, pass a different `--project`
 and `--dir`.
 **`... already has files in it; give --dir a new or empty directory`**
 The tool never writes over an existing deployment. Choose another directory, or
 destroy the old deployment first.
 **A deploy stopped partway.**
 The error says which step failed and shows the last lines of the relevant
 container's log. The stack is left as it was, for you to inspect. To start
 again from nothing: `ihasmail-oneshot destroy --dir DIR --yes`, fix the cause,
 and deploy again.
 **`Pool overlaps with other one on this address space`**
 The default private network `172.31.253.0/24` collides with a Docker network you
 already have. Destroy the partial deployment and deploy again with, for
 example, `--subnet 172.31.200.0/24`.
 **`Stalwart has no certificate yet`**
 Usually DNS: the mail host or one of the `autoconfig`, `autodiscover`,
 `mta-sts`, `ua-auto-config` names doesn't resolve to this host yet, or port 80
 isn't reachable from the internet. Check with `dig +short mail.example.com` from
 somewhere else, fix it, then run `ihasmail-oneshot certs --dir DIR`. Stalwart's
 reasons are in its log:
 ```bash
 docker compose logs stalwart | grep -i acme
 ```
 **`Caddy has no certificate for webmail.example.com yet`**
 Same causes. Caddy retries by itself with increasing delays. See
 `docker compose logs caddy | grep -i error`.
 **Mail arrives but sent mail never does.**
 Outbound port 25 is almost always the cause. Find the mail server of a domain
 you send to with `dig +short MX example.net`, then test from the host with
 `nc -vz <that mail server> 25`. Stalwart's log names each delivery
 failure (`docker compose logs stalwart`). Also check that reverse DNS for the host's address
 names the mail host.
 **Sent mail lands in spam.**
 Check that the SPF, DKIM and DMARC records from `dns-records.zone` are
 published exactly, and that reverse DNS is set. New addresses need a little time
 to build a sending reputation.
 **The webmail signs everyone out.**
 Expected after `docker compose restart ihasmail`, any ihasmail upgrade, or a
 reboot: sessions live in memory.
 **`/api/health` shows push accounts `pending` that never become `verified`.**
 Stalwart can't reach `https://webmail.example.com` from inside its container.
 Mail still updates in real time over the relay. The usual causes are a host
 firewall that drops traffic from Docker's bridge to the host's own public
 address, or the webmail name not resolving from the host.
 ## Known limits
 - **One mail domain per deploy.** More can be added afterwards in Stalwart or in
  ihasmail's Administration. Their certificates and DNS records are then yours
  to arrange.
 - **Linux Docker hosts only.** The tool drives the local Docker daemon and
  publishes ports on the host it runs on.
 - **IPv4 on the private network.** Ports are also published on IPv6 wherever
  Docker does so on the host.
 - **Fresh deployments only.** It doesn't import an existing Stalwart, or adopt a
  deployment it didn't write.
 - **Stalwart doesn't retry a failed certificate order** by itself, on a
  transient CA error either. `certs` starts a new one.
 - **Push by subscription isn't covered by the end-to-end test**, whose lab has no
  public DNS for Stalwart to resolve the webmail's name through. Its fallback,
  the relay, is what the test exercises.
 - Validated against **Stalwart 0.16.22**. Other Stalwart versions may change the
  registry objects the setup uses.
@@ -96,6 +96,16 @@ ok "deploy completed and signed in through ihasmail"
 grep -q "certificate  issued by CN=Pebble" "$WORK/deploy.log" || die "deploy did not report Stalwart's certificate"
 ok "deploy reported Stalwart's certificate"
 # No --ihasmail-image, so the deploy took the newest release -- and must have
 # written it down as a dated tag, never the moving :latest.
 image=$(sed -n 's/^    image: \(ghcr.io\/coffey-labs\/ihasmail[:@].*\)$/\1/p' "$DEPLOY/compose.yaml")
 [[ "$image" =~ ^ghcr\.io/coffey-labs/ihasmail:[0-9]{4}\.[0-9]{1,2}\.[0-9]{1,2}-(pr[0-9]+|g[0-9a-f]+)$ ]] \
  || die "compose.yaml does not pin ihasmail to a dated release: ${image:-none}"
 ok "compose.yaml pins ihasmail to $image"
 running=$(docker compose --project-directory "$DEPLOY" ps -q ihasmail | xargs docker inspect --format '{{range .Config.Env}}{{println .}}{{end}}' | sed -n 's/^IHASMAIL_VERSION=//p')
 [ "${running/+/-}" = "${image##*:}" ] || die "ihasmail runs $running, but compose.yaml says ${image##*:}"
 ok "the running ihasmail is the release compose.yaml names ($running)"
 # expect CODE curl-args...: retry for up to 30s until curl gets CODE. Caddy
 # obtains certificates for its names in parallel and in the background, so the
 # first handshake for any one of them can come a few seconds after deploy.
@@ -173,7 +183,7 @@ ok "dns-records.zone has the MX and DKIM records"
 certs_out=$("$BIN" certs --dir "$DEPLOY" 2>&1 || true)
 grep -q "already holds a certificate for $MAIL" <<<"$certs_out" || die "certs did not see the existing certificate: $certs_out"
-ok "certs recognises the certificate already issued"
+ok "certs recognizes the certificate already issued"
 # --- the auto-ban -------------------------------------------------------------
 # Last, so nothing earlier can be affected by a ban. Scans come from throwaway
@@ -22,16 +22,47 @@ import (
 	"strings"
 )
-// Versions this release was tested with, end to end. Stalwart is pinned
+// Stalwart is pinned to the release this version of the tool was tested with,
-// because ihasmail validates against one Stalwart release at a time; the
+// because a Stalwart upgrade migrates its store with no way back and ihasmail
-// ihasmail tag is the newest release at the time; Caddy is pinned so that a
+// validates against one Stalwart release at a time. Caddy is pinned so that a
-// redeploy months from now renders the same proxy.
+// deploy months from now renders the same proxy.
 //
 // ihasmail is not pinned here. A pin went stale within days of each release,
 // and ihasmail's image cleanup keeps ten releases, so an old default would in
 // time stop pulling at all. The default is the newest release instead, looked
 // up when the tool runs and written into compose.yaml as its dated tag -- see
 // deploy.ResolveIhasmail -- so what a deployment runs is still recorded and
 // nothing moves it afterwards.
 const (
 	DefaultStalwartImage = "stalwartlabs/stalwart:v0.16.22"
 	DefaultIhasmailImage = "ghcr.io/coffey-labs/ihasmail:2026.9.10-pr328"
 	DefaultCaddyImage    = "caddy:2.11.4"
 	IhasmailRepository = "ghcr.io/coffey-labs/ihasmail"
 	// NewestIhasmail is the default --ihasmail-image. Only full releases move
 	// this tag; prereleases never do.
 	NewestIhasmail = IhasmailRepository + ":latest"
 )
 // ihasmail's versions are the date of a commit and where it came from --
 // 2026.9.13+pr344, or 2026.9.13+g1fa6578 for a commit that arrived without a
 // pull request -- and its image tags are the same with the "+" as "-", since a
 // Docker tag may not contain "+".
 var ihasmailVersionRE = regexp.MustCompile(`^\d{4}\.\d{1,2}\.\d{1,2}\+(?:pr\d+|g[0-9a-f]{7,40})$`)
 // IhasmailTag is the image tag an ihasmail version is published under. A
 // version that is not a release's -- empty, or the 0.0.0 of a build nobody
 // gave a version to -- has none.
 func IhasmailTag(version string) (string, bool) {
 	if !ihasmailVersionRE.MatchString(version) {
 		return "", false
 	}
 	return strings.Replace(version, "+", "-", 1), true
 }
 // FollowsNewestIhasmail reports whether the plan still asks for the newest
 // ihasmail release rather than a particular image.
 func (p Plan) FollowsNewestIhasmail() bool { return p.IhasmailImage == NewestIhasmail }
 // Stalwart's ACME order covers these next to the mail host, all under the mail
 // domain: it is what its own DNS zone points at the mail host as CNAMEs, and
 // Caddy has to answer for every one of them on port 80 or the order fails.
@@ -123,7 +154,7 @@ var (
 	imageRE    = regexp.MustCompile(`^[a-z0-9][a-z0-9._/-]*(?::[A-Za-z0-9._-]+)?(?:@sha256:[a-f0-9]{64})?$`)
 )
-func normaliseHost(s string) string {
+func normalizeHost(s string) string {
 	return strings.TrimSuffix(strings.ToLower(strings.TrimSpace(s)), ".")
 }
@@ -136,7 +167,7 @@ func (o Options) Validate() (Plan, error) {
 	p := Plan{Local: o.Local}
-	p.Domain = normaliseHost(o.Domain)
+	p.Domain = normalizeHost(o.Domain)
 	switch {
 	case p.Domain == "" && o.Local:
 		p.Domain = "example.test"
@@ -153,7 +184,7 @@ func (o Options) Validate() (Plan, error) {
 		p.Domain = "domain.invalid"
 	}
-	p.MailHost = normaliseHost(o.MailHost)
+	p.MailHost = normalizeHost(o.MailHost)
 	if p.MailHost == "" {
 		p.MailHost = "mail." + p.Domain
 	}
@@ -164,7 +195,7 @@ func (o Options) Validate() (Plan, error) {
 		fail("--mail-host %q must be one label under the domain, e.g. mail.%s", o.MailHost, p.Domain)
 	}
-	p.WebmailHost = normaliseHost(o.WebmailHost)
+	p.WebmailHost = normalizeHost(o.WebmailHost)
 	if p.WebmailHost == "" {
 		p.WebmailHost = "webmail." + p.Domain
 	}
@@ -225,7 +256,7 @@ func (o Options) Validate() (Plan, error) {
 	}
 	p.StalwartImage = orDefault(o.StalwartImage, DefaultStalwartImage)
-	p.IhasmailImage = orDefault(o.IhasmailImage, DefaultIhasmailImage)
+	p.IhasmailImage = orDefault(o.IhasmailImage, NewestIhasmail)
 	p.CaddyImage = orDefault(o.CaddyImage, DefaultCaddyImage)
 	for flag, img := range map[string]string{"--stalwart-image": p.StalwartImage, "--ihasmail-image": p.IhasmailImage, "--caddy-image": p.CaddyImage} {
 		if !imageRE.MatchString(img) {
@@ -98,3 +98,41 @@ func TestEveryProblemAtOnce(t *testing.T) {
 		t.Errorf("got %d errors, want 3: %v", len(lines), err)
 	}
 }
 func TestIhasmailFollowsTheNewestReleaseUnlessNamed(t *testing.T) {
 	p, err := Options{Domain: "example.com"}.Validate()
 	if err != nil {
 		t.Fatal(err)
 	}
 	if !p.FollowsNewestIhasmail() || p.IhasmailImage != "ghcr.io/coffey-labs/ihasmail:latest" {
 		t.Errorf("default ihasmail image %q", p.IhasmailImage)
 	}
 	named, err := Options{Domain: "example.com", IhasmailImage: "ghcr.io/coffey-labs/ihasmail:2026.9.13-pr344"}.Validate()
 	if err != nil {
 		t.Fatal(err)
 	}
 	if named.FollowsNewestIhasmail() {
 		t.Error("a named image is treated as the newest release")
 	}
 	byDigest := "ghcr.io/coffey-labs/ihasmail@sha256:" + strings.Repeat("a", 64)
 	if _, err := (Options{Domain: "example.com", IhasmailImage: byDigest}).Validate(); err != nil {
 		t.Errorf("a by-digest image is refused: %v", err)
 	}
 }
 func TestIhasmailTag(t *testing.T) {
 	for version, want := range map[string]string{
 		"2026.9.13+pr344":    "2026.9.13-pr344",
 		"2026.10.2+g1fa6578": "2026.10.2-g1fa6578",
 		"0.0.0":              "",
 		"":                   "",
 		"2026.9.13":          "",
 		"2026.9.13+pr344\n":  "",
 		"latest":             "",
 	} {
 		got, ok := IhasmailTag(version)
 		if got != want || ok != (want != "") {
 			t.Errorf("IhasmailTag(%q) = %q, %v; want %q", version, got, ok, want)
 		}
 	}
 }
@@ -84,6 +84,50 @@ func Preflight(ctx context.Context, p config.Plan, log Log) (warnings []string,
 	return warnings, errors.Join(problems...)
 }
 // ResolveIhasmail turns "the newest ihasmail release" into the image that is
 // the newest release right now, so compose.yaml records a version rather than
 // a tag that moves. A plan that names its own image is returned as it is.
 //
 // The dated tag is taken from the version the image itself carries, and used
 // only once the registry confirms that tag is the very same image; otherwise
 // the image is pinned by digest, which is exact but says less to a person
 // reading compose.yaml. Either way a later `docker compose pull` cannot move
 // the deployment onto a release nobody chose.
 func ResolveIhasmail(ctx context.Context, p config.Plan, log Log) (config.Plan, error) {
 	if !p.FollowsNewestIhasmail() {
 		return p, nil
 	}
 	log.Step("finding ihasmail's newest release")
 	if err := docker.Pull(ctx, config.NewestIhasmail); err != nil {
 		return p, fmt.Errorf("could not fetch ihasmail's newest release (%w); name an image with --ihasmail-image to use another", err)
 	}
 	newest, err := docker.ImageID(ctx, config.NewestIhasmail)
 	if err != nil {
 		return p, err
 	}
 	version, err := docker.ImageEnv(ctx, config.NewestIhasmail, "IHASMAIL_VERSION")
 	if err != nil {
 		return p, err
 	}
 	if tag, ok := config.IhasmailTag(version); ok {
 		dated := config.IhasmailRepository + ":" + tag
 		if docker.Pull(ctx, dated) == nil {
 			if id, err := docker.ImageID(ctx, dated); err == nil && id == newest {
 				p.IhasmailImage = dated
 				log.Info("ihasmail  %s, recorded as %s", version, dated)
 				return p, nil
 			}
 		}
 	}
 	digest, err := docker.RepoDigest(ctx, config.NewestIhasmail, config.IhasmailRepository)
 	if err != nil {
 		return p, err
 	}
 	log.Warn("ihasmail's newest release (version %q) has no matching dated tag; recording it by digest", version)
 	p.IhasmailImage = digest
 	return p, nil
 }
 // portFree tries to bind an address. A permission error means an unprivileged
 // user asking about a low port, which says nothing about whether Docker can
 // have it, so it is not reported.
@@ -33,6 +33,54 @@ func Output(ctx context.Context, args ...string) (string, error) {
 	return strings.TrimSpace(stdout.String()), nil
 }
 // Pull pulls one image. Quiet, because it runs before the plan is confirmed
 // and the step already says what it is fetching.
 func Pull(ctx context.Context, image string) error {
 	_, err := Output(ctx, "pull", "--quiet", image)
 	return err
 }
 // ImageID is the local ID of an image, which is the same for two references
 // only when they are the same image.
 func ImageID(ctx context.Context, image string) (string, error) {
 	return Output(ctx, "image", "inspect", "--format", "{{.Id}}", image)
 }
 // ImageEnv is the value an image's configuration gives an environment
 // variable, or "" when it sets none.
 func ImageEnv(ctx context.Context, image, name string) (string, error) {
 	out, err := Output(ctx, "image", "inspect", "--format", "{{range .Config.Env}}{{println .}}{{end}}", image)
 	if err != nil {
 		return "", err
 	}
 	return envValue(out, name), nil
 }
 func envValue(env, name string) string {
 	for _, line := range strings.Split(env, "\n") {
 		if v, ok := strings.CutPrefix(line, name+"="); ok {
 			return v
 		}
 	}
 	return ""
 }
 // RepoDigest is an image's by-digest reference in one repository, e.g.
 // ghcr.io/coffey-labs/ihasmail@sha256:..., which names exactly that image for
 // as long as the registry keeps it.
 func RepoDigest(ctx context.Context, image, repository string) (string, error) {
 	out, err := Output(ctx, "image", "inspect", "--format", "{{range .RepoDigests}}{{println .}}{{end}}", image)
 	if err != nil {
 		return "", err
 	}
 	for _, d := range strings.Fields(out) {
 		if strings.HasPrefix(d, repository+"@") {
 			return d, nil
 		}
 	}
 	return "", fmt.Errorf("%s has no digest from %s", image, repository)
 }
 // Versions returns the engine and compose versions, which is also the check
 // that both are installed and this user may use them.
 func Versions(ctx context.Context) (engine, compose string, err error) {
@@ -48,7 +96,7 @@ func Versions(ctx context.Context) (engine, compose string, err error) {
 	return engine, compose, nil
 }
-// ProjectLeftovers lists containers, volumes and networks already labelled
+// ProjectLeftovers lists containers, volumes and networks already labeled
 // with a compose project name. Any at all means an earlier run of the same
 // project, and its volumes would hand a "fresh" deployment an old server.
 func ProjectLeftovers(ctx context.Context, project string) ([]string, error) {
@@ -0,0 +1,21 @@
 // SPDX-FileCopyrightText: 2026 Coffey Labs
 // SPDX-License-Identifier: AGPL-3.0-or-later
 package docker
 import "testing"
 func TestEnvValue(t *testing.T) {
 	env := "PATH=/usr/local/bin:/usr/bin\nNODE_ENV=production\nIHASMAIL_VERSION=2026.9.13+pr344\nEMPTY=\n"
 	for name, want := range map[string]string{
 		"IHASMAIL_VERSION": "2026.9.13+pr344",
 		"NODE_ENV":         "production",
 		"EMPTY":            "",
 		"MISSING":          "",
 		"IHASMAIL":         "", // a prefix of a name is not the name
 	} {
 		if got := envValue(env, name); got != want {
 			t.Errorf("envValue(%q) = %q, want %q", name, got, want)
 		}
 	}
 }
@@ -31,6 +31,8 @@ services:
        ipv4_address: {{.Plan.StalwartIP}}
  ihasmail:
    # A fixed release: `docker compose pull` never moves it. To upgrade, change
    # the tag here, then `docker compose pull && docker compose up -d`.
    image: {{.Plan.IhasmailImage}}
    restart: unless-stopped
    depends_on: [stalwart]
Author	SHA1	Message	Date
jcoffey-dev	7d3c4d4524	ci: add Gitea Actions workflow ported from .gitlab-ci.yml ci / test (pull_request) Successful in 35s Details ci / release (pull_request) Skipped Details ci / test (push) Successful in 1m19s Details ci / release (push) Failing after 16s Details	2026-09-21 22:44:22 -07:00
jcoffey-dev	153077eab9	Merge branch 'ci/release-permalink' into 'main' Give releases a stable latest-download URL See merge request coffey-labs/ihasmail-oneshot!2	2026-09-20 21:15:14 -07:00
jcoffey-dev	a4f0953353	Give releases a stable latest-download URL The install guide tells people to curl .../releases/latest/download/<file> which is a GitHub URL shape. GitLab's equivalent is /-/releases/permalink/latest/downloads/<path> but it only resolves for assets that declare direct_asset_path, and the release job was creating plain links to the package registry. Those carry the tag in the URL, so they can never be a "latest" link. Each asset now also declares /binaries/<file>, which is what the docs will point at. The path is load-bearing: changing it breaks a documented install command.	2026-09-20 20:58:45 -07:00
jcoffey-dev	4627b95ac9	Merge branch 'ci/gitlab-pipeline' into 'main' Run CI on the self-hosted GitLab See merge request coffey-labs/ihasmail-oneshot!1	2026-09-20 20:35:35 -07:00
jcoffey-dev	2d64e30cb2	Run releases on the self-hosted GitLab Ports .github/workflows/release.yml after the GitHub account was suspended: tag-driven, reproducible tarballs, the same refusal to release a tag that is not an ancestor of the default branch, with the assets going to the generic package registry and a Release created from them. e2e.yml is not ported. e2e/public.sh publishes 25, 80, 443, 465, 993, 995 and 4190 on the machine it runs on. On Actions that was a throwaway VM; the runner here is Web_Host, where 80 and 443 are nginx serving every live site. It stays a manual check on a disposable host. The Actions workflows stay in the tree as the reference.	2026-09-20 20:13:35 -07:00
jcoffey	f874cb8651	Merge pull request #8 from Coffey-Labs/docs/guide-link-first README: documentation links first	2026-09-15 12:31:19 -07:00
jcoffey-dev	bbd84da276	Put the documentation links first, with the guide marked as the place to start	2026-09-15 12:25:35 -07:00
jcoffey	07ed6ea582	Merge pull request #7 from Coffey-Labs/docs/shorter-readme Shorter README; technical detail moves to docs/	2026-09-15 12:16:51 -07:00
jcoffey-dev	97995fdebb	Shorten the README; move the technical detail into docs/ The README keeps what the tool is, how to install it and the first commands, and points to the guide on docs.ihasmail.org. Everything else moves, whole, into docs/ and CONTRIBUTING.md, where it is organized for readers who want the detail. Where the old README disagreed with the code, the code wins.	2026-09-15 12:14:34 -07:00
jcoffey	74fa9321f9	Merge pull request #6 from Coffey-Labs/feat/resolve-latest-ihasmail Deploy ihasmail's newest release, recorded by its dated tag	2026-09-15 12:01:48 -07:00
jcoffey-dev	408fc20d7b	Deploy ihasmail's newest release, recorded by its dated tag The ihasmail default was a pin that went stale within days, and ihasmail keeps ten releases' images, so an old default would in time stop pulling. With no --ihasmail-image the tool now pulls :latest before asking, reads the version the image carries, confirms the dated tag is the same image, and writes that tag into compose.yaml (or the digest, if there is no such tag). Stalwart and Caddy stay pinned. A weekly end-to-end run against the newest release, three hours after ihasmail publishes, is what keeps it safe.	2026-09-15 12:00:20 -07:00
jcoffey	c151df45d0	Merge pull request #5 from Coffey-Labs/chore/us-spelling-license Use American English spelling	2026-09-15 11:47:13 -07:00
jcoffey-dev	eb1ccc415e	Use American English spelling throughout	2026-09-15 11:46:08 -07:00
jcoffey-dev	1765b31d5e	Spell license the US way	2026-09-15 11:35:45 -07:00
jcoffey	c6e1d026d1	Merge pull request #4 from Coffey-Labs/docs/drop-gpl-release-note Drop the README note about the GPL release	2026-09-13 22:39:24 -07:00
jcoffey-dev	83ba353b9f	Drop the README note about the GPL release v2026.9.13 has been deleted, so the note named a release nobody can find.	2026-09-13 22:39:19 -07:00