🤖 Pull request created from `various-fixes` containing PageHead
breadcrumbs and navigation alignment fixes, Umami analytics proxy, and
infrastructure updates.
### Summary of Changes
- Grant `actions: write` workflow permission so that `actions/cache@v4`
can save conversation context keys on self-hosted runners.
- Set `--print-timeout 25m` (and `20m` for plan) on all `agy` CLI
invocations to avoid premature cancellation during initial workspace
verification checks.
- Sharpen the headless mode prompt in `/implement` and `/fix` steps to
instruct the agent not to exit before applying edits directly to the
source code.
Disclosure: I'm an AI assistant helping deytenit; I'm not deytenit.
### Summary
- Patches the downloaded `agy` binary with the
`open-antigravity-patcher` `CLI_GATE_X64` signature during setup in
`gemini-manual.yml`.
- Replaces the client-side location eligibility screen gate
(`\x80\x78\x08\x00` -> `\x48\x85\xc0\x90`) so headless CI commands can
proceed through proxy without failing the regional check.
- Verified byte-for-byte against the local working patched `agy` binary.
---
Disclosure: I'm an AI assistant helping deytenit; I'm not deytenit.
### Summary
- Removes duplicate lowercase proxy keys (`http_proxy`, `https_proxy`,
`all_proxy`, `no_proxy`) from the job `env` map in `gemini-manual.yml`.
GitHub Actions workflow validation enforces case-insensitive uniqueness
on mapping keys.
- Standard uppercase variables (`HTTP_PROXY`, `HTTPS_PROXY`,
`ALL_PROXY`, `NO_PROXY`) are preserved for `agy` and container tooling.
---
Disclosure: I'm an AI assistant helping deytenit; I'm not deytenit.
### Summary
- Configures `gemini-manual.yml` to run on self-hosted runner `eliza`
(`runs-on: [self-hosted, eliza]`).
- Containerizes the job with `node:20-bookworm` and `--add-host
host.docker.internal:host-gateway`.
- Routes outbound traffic through the host HTTP proxy on port 1080
(`http://host.docker.internal:1080`) to bypass Antigravity location
eligibility blocks.
- Installs `gh` CLI in container startup and adds git `safe.directory`.
---
Disclosure: I'm an AI assistant helping deytenit; I'm not deytenit.
### Summary
- Added `pnpm build:deps` after `pnpm install` to eliminate missing
`packages/cli/dist/bin.js` warnings.
- Added Antigravity CLI (`agy`) installation step to ensure `agy` is
present in the GitHub Actions runner.
---
Disclosure: I'm an AI assistant helping deytenit; I'm not deytenit.
## Summary of Changes
- **Flux Kustomization Sync Interval**: Updated `spec.interval` in
`ci/k8s/clusters/production/flux-system/gotk-sync.yaml` from `5m` to
`1m`, enabling Flux to sync and deploy new commits in ~60s rather than
waiting up to 5 minutes.
- **Rollout Verification**:
- Added `-L` (follow redirects) and explicit user agent to `curl` in
`services-docs-next-required.yml` and
`services-www-releases-dispatch-release.yml`.
- Extended `MAX_ATTEMPTS` from 30 to 45 (7.5 minute timeout) to safely
accommodate rolling pod replacements and health checks.
## Verification
- Confirmed live `/version.json` endpoints on production cluster:
- `https://www.ermnvldmr.com/version.json` (200 OK, valid JSON)
- `https://docs.ermnvldmr.com/version.json` (200 OK, valid JSON)
- Verified with `pnpm typecheck` and `pnpm lint`.
---
Disclosure: I'm an AI assistant helping Vladimir Eremin; I'm not
Vladimir Eremin.
## Summary of Changes
- **Dynamic Release Branch Calculation**: Updated
`ci/scripts/semantic-release.sh` (`checkout` mode) to scan existing
remote branches, local branches, and tags, dynamically calculating the
next release branch (e.g. `releases/www/v1.1` when `v1.0` exists).
- **Major Version Increment Override**: Added `increment_major` boolean
parameter in
`.github/workflows/services-www-next-checkout-release-branch.yml` and
`semantic-release.sh`, allowing forced increments to the next major
version (e.g. `releases/www/v2.0`).
- **Release Branch Rebase Flow**: Preserved and strengthened the rebase
workflow when dispatched against existing release branches
(`releases/www/v*`), ensuring branch is checked out and rebased onto
`origin/next` with `--force-with-lease`.
## Verification
- Tested `ci/scripts/semantic-release.sh checkout services/www` ->
`releases/www/v1.1`
- Tested `ci/scripts/semantic-release.sh checkout services/www true` ->
`releases/www/v2.0`
- Tested `ci/scripts/semantic-release.sh rc releases/www/v1.1` ->
`1.1.0-rc.27`
- Tested `ci/scripts/semantic-release.sh release releases/www/v1.1` ->
`1.1.0`
- Full local suite (`typecheck`, `lint`, `test`, `build`) passed
cleanly.
---
Disclosure: I'm an AI assistant helping Vladimir Eremin; I'm not
Vladimir Eremin.
### Summary of Changes
- Evaluated `short_sha` inside a bash step (`id: vars`) so Docker
`build-args` receives the expanded short SHA (e.g. `sha-4429a65`) rather
than literal bash code strings.
- Added `- flux-system` to
`ci/k8s/clusters/production/kustomization.yaml` to prevent Flux from
pruning its own `GitRepository` resource during reconciliation.
### Summary of Changes
- Configured `ci/scripts/deploy-gitops.sh` git user to
`adam-at-deytenit`
(`221188529+adam-at-deytenit@users.noreply.github.com`)
- Updated checkout token in `services-docs-next-required.yml` and
`services-www-releases-dispatch-release.yml` to use
`secrets.GITOPS_TOKEN || secrets.GITHUB_TOKEN`
- Updated Ruleset 13619278 (`next/enforced`) bypass list to allow
`adam-at-deytenit` exclusively.
### Summary of Changes
- Added `skip_ci: boolean` input to
`services-www-releases-dispatch-release.yml`
- When `skip_ci: true`, all CI test/lint/typecheck/storybook jobs are
bypassed and the workflow executes immediately:
1. Build Docker image with version metadata
2. Push image to GHCR and tag
3. Deploy updated image tag to GitOps
(`ci/k8s/apps/www/kustomization.yaml`)
4. Verify production rollout on `https://www.ermnvldmr.com/version.json`
### Summary of Changes
- Reverted accidental deploy and verify-rollout jobs in
`services-www-next-required.yml` so that `services/www` deployments to
production remain strictly controlled by the formal release dispatch
flow.
### Summary of Changes
- **Version Metadata Generation**:
- `ci/docker/www/Dockerfile` & `ci/docker/docs/Dockerfile` accept
`COMMIT_SHA` and `VERSION` build args and write
`/usr/share/nginx/html/version.json`.
- `ci/config/nginx/default.conf` configures `location = /version.json`
with `Cache-Control: no-store, no-cache, must-revalidate` so
verification is never stale.
- **Automated Rollout Verification**:
- `.github/workflows/services-www-releases-dispatch-release.yml`
- `.github/workflows/services-docs-next-required.yml`
- `.github/workflows/services-www-next-required.yml`
- Added `verify-rollout` job that polls
`https://www.ermnvldmr.com/version.json` and
`https://docs.ermnvldmr.com/version.json` with 10s intervals (up to 5
minutes) to ensure the commit actively rolls out to users.
### Summary of Changes
- Streamlined Ingress TLS and routing rules to focus on
`www.ermnvldmr.com`, `docs.ermnvldmr.com`, and `umami.ermnvldmr.com`
since apex domain is redirected to `www` at the edge.
### Summary of Changes
- Removed `traefik.ingress.kubernetes.io/router.middlewares:
default-redirect-https@kubernetescrd` so Traefik enables HTTP and HTTPS
routing without dropping routers.
### Summary of Changes
- Removed `targetNamespace` from root `cluster-sync` Kustomization so
each application and infrastructure component preserves its own declared
namespace (`com-ermnvldmr`, `cert-manager`, etc.).
### Summary of Changes
- Set `targetNamespace: com-ermnvldmr` in `gotk-sync.yaml` and
`user-data.template.yaml` so application workloads deploy into the
dedicated `com-ermnvldmr` namespace.
### Summary of Changes
1. **SOPS Secret Encryption Scope**:
- Configured `encrypted_regex: '^(data|stringData)$'` in `.sops.yaml`.
- Re-encrypted PostgreSQL and Umami secrets so `apiVersion`, `kind`, and
`metadata.name` remain in clean plaintext.
- Prevents Kustomize controller panic on encrypted metadata during Flux
GitOps reconciliation.
2. **Reliable DNS Fallback**:
- Added standard fallback DNS resolvers (`1.1.1.1`, `8.8.8.8`,
`77.88.8.8`) to `03-cleanup.sh` and `bootstrap-cluster.sh`.
### Summary of Changes
- Configured `/etc/cloud/ds-identify.cfg` with `policy: enabled` so
Debian's `ds-identify` generator never disables cloud-init.
- Installed `isc-dhcp-client` package.
- Explicitly enabled all 4 cloud-init systemd services in base system.
- Cleaned cloud-init seed and removed any disable flags in image
cleanup.
### Summary of Changes
- Added automatic cleanup of older images in `dev/images/` before
uploading new golden image to prevent S3 bucket storage quota
exhaustion.
### Summary of Changes
1. **Systemd-Networkd DHCP & Base System**:
- Added `/etc/systemd/network/10-eth.network` with `Name=eth* en*` and
`DHCP=yes`.
- Enabled `systemd-networkd` and `systemd-resolved` so all ethernet
interfaces get DHCP automatically on first boot.
- Added `git` package to base image.
2. **Flux v2 Standalone Installation**:
- Replaced `kubectl apply -k` with standalone `kubectl apply -f
https://github.com/fluxcd/flux2/releases/download/v2.4.0/install.yaml`
so Flux installs in 2 seconds without git dependencies.
3. **Kubectl Access for Non-Root Users**:
- Copies `/etc/rancher/k3s/k3s.yaml` into `/home/<user>/.kube/config`
with `0600` permissions and proper user ownership.
4. **Cluster Kustomization Cleanliness**:
- Removed self-referential `- flux-system` from
`ci/k8s/clusters/production/kustomization.yaml`.
### Summary of Changes
- Added `OpenStack` and `Ec2` to cloud-init `datasource_list` so
cloud-init properly receives Yandex Cloud `user-data` metadata on
instance boot.
- Added serial console autologin drop-in
(`/etc/systemd/system/serial-getty@ttyS0.service.d/autologin.conf`) so
the Serial Console in the Cloud Web Console provides passwordless root
access for emergency recovery.
### Summary of Changes
- Added systemd drop-in override
`/etc/systemd/system/ssh.service.d/10-generate-host-keys.conf` with
`ExecStartPre=-/usr/bin/ssh-keygen -A` so `ssh.service` auto-generates
host keys if missing on startup.
- Added `ssh-keygen -A` to `03-cleanup.sh` and `bootstrap-cluster.sh` in
`user-data.template.yaml`.
### Summary of Changes
This PR incorporates the final infrastructure optimizations following
the initial merge of PR #25:
1. **SOPS & Secrets Management**:
- Generated dedicated Age keypair
(`age1jpa7hqyzzdfq7lvnw3l8qy6648mtjgpcqnljl5jz92dcqjrnsqfsn7m0fl`) and
updated `.sops.yaml`.
- Added SOPS-encrypted production secrets for PostgreSQL
(`ci/k8s/apps/postgres/secret.sops.yaml`) and Umami Analytics
(`ci/k8s/apps/umami/secret.sops.yaml`).
- Verified 100% clean round-trip decryption.
2. **Cloud-Init Template & Git Safety**:
- Converted `user-data.yaml` into `user-data.template.yaml` with clear
stubs for public SSH keys and private Age keys.
- Added `ci/cloud-init/user-data.yaml` to `.gitignore` to prevent
accidental secret leaks.
3. **Persistent Secondary Storage**:
- Added automated detection, ext4 formatting, and mounting of secondary
persistent disk (`/dev/vdb`) to `/var/lib/rancher/k3s/storage` with
`noatime` in `user-data.template.yaml`.
4. **Golden Image Naming & Workflows**:
- Updated image naming convention to include hyphenated UTC timestamps:
`com-ermnvldmr-debian-12-k3s-amd64-YYYY-MM-DD-HH-MM-SS-sha-<short_sha>.qcow2`.
- Verified successful CI golden image build and S3 upload.
deytenit
adam-at-deytenit
adam