[gabemeola]

Summary

• Make OpenCode v2.2.4 the default in version.txt; keep supported v1.19.5 in version-v1.txt. The scheduled checker opens independent bump PRs for both major lines.
• Publish both pinned tracks with task publish. V2 receives its version tag, :latest, and :v2; v1 receives its version tag and :v1. Existing GitHub releases are retained.
• Use the correct upstream installer and select the major-specific entrypoint in the Dockerfile: entrypoint.sh for v2, entrypoint-v1.sh for v1. Both share home/tool initialization.
• Support v2's native OPENCODE_PASSWORD and legacy OPENCODE_SERVER_PASSWORD fallback. Preserve generated-password behavior when neither is set. V2 uses the fixed username opencode; v1 retains OPENCODE_SERVER_USERNAME.
• Normalize v2 log levels to lowercase and enable its stderr logging correctly. Keep foreground serve with explicit hostname, port, and CORS flags.
• Test the actual API routes: /api/info for v2 and /global/health for v1. Old v1 API URLs on v2 return the web app and cannot validate API authentication.
• Run runtime CI for both versions, update the home-tool tests for v2's API, and build the size-comparison baseline using its own pinned version.

Local verification

• Full Docker builds passed for v2.2.4 and v1.19.5 on linux/arm64.
• Server E2E passed for both images: build-time entrypoint selection, custom port and CORS, uppercase log-level compatibility, stderr logs, real JSON API responses, missing/incorrect credentials, username behavior, and graceful shutdown.
• V2 additionally verified OPENCODE_PASSWORD precedence, OPENCODE_SERVER_PASSWORD fallback, generated-password auth, and managed mise instructions.
• Home-tool installation E2E passed for both images.
• All seven bump-script tests passed; shell and workflow/Taskfile YAML checks passed; task publish dry run confirms all image tags.

One Sapling commit; no images or releases published by these checks.

Review follow-up

• Release discovery now uses authenticated, paginated GitHub API requests, selects the highest numeric stable version per major, excludes component/prerelease/draft tags, and never downgrades the pins. The scheduled checker and task update share this implementation.
• GitHub release creation explicitly targets the checked-out source commit. V2 is marked Latest; v1 uses --latest=false. Repository references use the canonical sprisa/opencode-server name.
• Release notes select the highest older stable tag reachable from the current commit, including the first-v2 fallback to a lower major, while excluding future versions and unmerged tags.
• Added eight offline release-automation integration tests and a dedicated CI job. All eight tests and the seven existing bump tests pass locally. Live upstream lookup returns v2.2.4 and v1.19.5.
• Verified a real persistent-volume upgrade locally: create a v1 session, stop v1, start v2 on that same volume as UID 0, and retrieve the session with the same ID and title through the v2 API.
• README links the upstream v1 migration guide for changed client and plugin APIs.

Still one Sapling commit.

Toolchain refresh and verification

• Homebrew: 6.0.16 -> 7.0.11.
• mise: 2026.8.3 -> 2026.10.24.
• zerobrew: 0.3.2 -> 0.4.2; download URL uses the canonical zerobrewhq repository.
• mise-zerobrew: 998174989a12910f82c1f6e791a432731ed2647d is already the latest upstream commit and remains pinned.
• Fixed new mise compatibility: the scoped home-tool installer now uses an actual empty TOML config instead of /dev/null, which current mise rejects as an unknown config type.
• Fixed readiness polling to wait through v2's temporary startup 503. Added progress messages, bounded HTTP requests, and a five-minute timeout for the package-manager E2E test.
• Full local builds passed for all four combinations: OpenCode v1/v2 on linux/arm64 and linux/amd64.
• Both major versions on both architectures passed server authentication/flags, delayed home-tool installation, Ruby/Wget/Nano lazy installs through the zerobrew plugin, a real Homebrew package install, and zbx execution.
• Interactive and non-interactive Node/npm/npx/Python/GitHub CLI/Go checks also passed on arm64 and amd64. Added package-manager E2E coverage to both runtime CI jobs.

All changes remain in the single Sapling commit.

[github-actions]

Docker image size comparison

Image	Size
Baseline (main)	989MB
This PR	1.17GB
Difference	+66MB