CloudSeed v2.2.0-rc.6

Fixed

• Exit code on failure: every command now exits 1 when it fails. Previously generate/validate/completion printed the error but exited 0, so CI could not detect a failure.
• --fix-it no longer destroys the password: it now writes a real SHA-512 crypt hash ($6$...) with a random salt, using the password_rounds from the config. It previously wrote the literal placeholder $6$rounds=5000$examplesalt$, which was not a hash of anything and locked the operator out of the target.
• Documented platforms now work: vsphere, kvm and physical are accepted. The provider previously only accepted local, aws, azure and gcp, so every configuration in the documentation failed with "Unsupported platform".
• Config format is decided by the file extension, not guessed by trying YAML and falling back to TOML. A malformed file now reports the format it tried, and an unknown extension is rejected instead of being silently parsed as TOML.
• Config fields are no longer silently dropped: Config now models the fields the documentation describes. Serde used to ignore all of them, so a documented config generated output that ignored almost everything.
• The generated cloud-config uses the real configuration: user name, SSH keys, groups, sudo, packages, timezone, locale, network and files are all rendered. The template previously hardcoded the user ubuntu and a truncated, invalid SSH key (ssh-rsa AAAAB3NzaC1yc2EAAAADAQABAAABAQ...) that ended up in every generated VM.
• The modules list is honoured: it was parsed and then never read by any provider.
• Output layout matches the documentation: files are written to out/<platform>-<os>/ with user-data, meta-data, cloudseed.yaml and README.txt (plus the Cloudbase-Init and Sysprep artifacts on Windows). Previously a single user-data.yaml was written into the output directory.
• Static networking is emitted as cloud-init network config version 2, converting the dotted netmask to a prefix length.
• --json-logs actually emits JSON instead of printing "not implemented" and falling back to plain text.
• validate accepts a directory of configurations, as its help text claimed.
• TUI and CLI agree on the supported shells: the TUI offered only bash, zsh and fish while the CLI also supported powershell and elvish.
• Ctrl+C quits the TUI from any screen. Previously only q on the menu bar worked, so leaving a form took several Esc presses.
• A field whose module is disabled is reported: setting packages while the packages module is off used to be silently dropped. The validator now names the field and the module.
• An unset timezone is no longer forced to UTC: the timezone has no default now, so the generated cloud-config only carries a timezone key when one was configured, and the image default is left alone.
• The documented config examples are valid again: the README's main example was missing version and name and carried a plaintext password, and a ### heading had been left inside the YAML code block, which broke the example.
• --version works: the flag was documented but the command had no version, so it was rejected as an unknown argument.
• --quiet actually quiets the output: it only removed the log line numbers. It now drops the log level to warnings and errors.

Added

• cloudseed detect-cloud-init: reports the local cloud-init version and whether CloudSeed supports it.
• cloudseed write-to-cloud-init-path: writes the generated config to /etc/cloud/cloud.cfg.d/99-cloudseed.cfg on a running Linux host.
• Windows artifacts: cloudbase-init.conf, cloudbase-init-unattend.conf, sysprep-unattend.xml and run-sysprep.bat are now generated, and the cloud-config is written as user-data for Cloudbase-Init to consume.
• vSphere Guest Customization Specification export (vsphere-customization-spec.xml) and the pre/post customization scripts.
• Validation rules for unknown platforms, unknown modules, unknown OS types, static networking without an address or gateway, a locked password without an SSH key, and a Windows target without Sysprep.
• network_domain, sysprep_* and vsphere_* configuration fields, all of which were previously documented but ignored.
• template_best_practices is now implemented: it writes a template-cleanup.sh / template-cleanup.bat golden-image cleanup script and explains in README.txt how to seal the image. It is opt-in, because it adds a file.

Changed

• The release workflow now handles pre-release tags: a tag with a suffix (v2.2.0-rc.1) publishes the release as a pre-release, the changelog section is looked up by the plain version (2.2.0), so the release body is no longer empty, and the release is not marked as the latest one, which the API rejects for a pre-release.
• The changelog lookup no longer depends on the awk implementation: it uses a literal substring match, which behaves the same under gawk and mawk. The previous regex was passed to awk through a -v assignment, and gawk processes escape sequences there, so the escaped pattern became a character class that never matched. The runners use gawk, so the release body came out empty there and correct locally.
• Releases are created through the GitHub API directly, with an explicit payload. The release action creates a pre-release as a draft and publishes it in a second call, and that second call was what the API rejected, so the release was never created. The workflow now posts the payload it wants and uploads the assets with gh release upload.
• scripts/simulate-release.sh dry-runs the release metadata: it extracts the real step scripts from the workflow, runs them against the real changelog, and validates the resulting API payload, so the version, the pre-release flag, the latest flag and the release body can be checked before anything is published.
• One workflow publishes releases, not two: the duplicate release job in ci.yml published raw binaries alongside the ZIPs and could overwrite the release. release.yml is now the only publisher, and it produces the Linux and Windows ZIPs.
• Password hashing uses sha-crypt, adding one dependency. The handlebars dependency was dropped: the cloud-config is now built as typed values and serialized with serde_yaml, which handles quoting and block scalars correctly.
• A test guards the documentation: every yaml block in README.md and GUIDE.md must parse as a valid config and pass validation, so the examples cannot drift away from the code again.
• The cloudseed-testutils crate was removed; it was empty.
• The aws/azure/gcp feature flags were removed from cloudseed-core; they were never referenced. The provider surface is now a single PlatformProvider parameterised by platform and OS.
• Shared command functions (cmd_generate, cmd_validate, cmd_completion, set_timezone) are used by both the CLI and the TUI, so the two surfaces cannot drift.

Removed

• cloudseed-core feature flags aws, azure, gcp (dead).
• Dead render_template helper in cloudseed-templates.