HHostingByteSoftware
Ctrl K
ComunidadGuías…

CHANGELOG.md

Rama main · réplica de lectura del repositorio original.

Archivo original ↗Preguntar a KiBot
# Changelog

All notable changes to this project will be documented in this file.

The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).

## [0.11.0] - 2026-09-18

Security-focused round, prompted by an external code review of `internal/cloudflare`, `internal/cloudflared`, `config`, and `cmd/zt/init.go`.

### Added

- `internal/validate.TunnelName` — a single, central check for what counts as a valid tunnel/service name (`^[a-z0-9][a-z0-9-]{0,62}$`), applied everywhere a name is accepted: `zt up`, `zt down`, `zt restart`, `zt status`, `zt logs`, and `zt.yaml` manifest loading. Previously a raw, unvalidated name was used directly as a filesystem directory (`~/.zt/tunnels/<name>`), a systemd/launchd/Task Scheduler unit name, and a DNS hostname prefix all at once, with nothing stopping a value like `../foo` or `foo/../../bar` from escaping the intended directory that `zt down`'s `os.RemoveAll` operates on
- `zt.yaml` service names are now validated at `Load` time (same `TunnelName` check), so `zt apply`/`zt export` catch an invalid name before it's used to create Cloudflare resources
- `ZT_API_TOKEN`, `ZT_ACCOUNT_ID`, `ZT_DOMAIN` environment variables, as an alternative credential source to `~/.zt-config.json`. Each one, if set, overrides the corresponding value from the config file; set all three and `zt init` isn't needed at all — useful on ephemeral CI runners/containers, or to rotate the token without re-running `zt init` interactively
- Every tunnel `zt up`/`zt apply` creates is now tagged with Cloudflare Tunnel metadata `managed_by: cfzt` (`internal/cloudflare.FindTunnelByNameManaged`)

### Changed

- **Breaking:** `zt up`/`zt apply` no longer treats any same-name Cloudflare tunnel as automatically stale and safe to delete. A tunnel tagged `managed_by: cfzt` (cfzt's own, e.g. left behind by an interrupted run) is still deleted and recreated automatically. A same-name tunnel *without* the tag — created via the Cloudflare dashboard, another tool, or a pre-0.11.0 cfzt release — now makes `zt up` refuse and exit with an error instead of silently deleting it; pass `--force` to delete it anyway. Tunnels created before this release aren't tagged, so the first `zt up` that collides with one of them will need `--force`
- `--force` on `zt up`/`zt apply` now also prints an explicit warning (record type + what it was pointing to) whenever it actually deletes a foreign DNS record, instead of doing so silently. Still a plain printed warning, not a blocking confirmation prompt — same reasoning as `--public`'s warning: a prompt would break `cfzt-action`'s non-interactive PR-preview workflow. Grep your CI logs for it if you're not watching a run live
- `zt init` no longer echoes the API token to the terminal while typing — hidden via `golang.org/x/term.ReadPassword` on a real TTY, with a transparent fallback to a normal line read when stdin is redirected (scripted input, or the existing tests)

## [0.10.2] - 2026-08-26

### Added

- `zt apply` now remembers the content of the manifest it last applied and detects if `zt.yaml` changed on disk since then — before, `zt apply` trusted whatever manifest file it was pointed at with no way to notice an unexpected edit. On a mismatch it prints a diff and refuses to proceed until re-run with the new `--force` flag

## [0.10.1] - 2026-08-09

### Added

- `--container-port` flag for `zt up --docker` (and matching `container_port` field in `zt.yaml` manifests) to pin which published port to expose when a container publishes more than one — previously only the lowest port was reachable

### Changed

- README split: full details for the GitHub Action, Windows support, backup/restore, the watchdog, troubleshooting, and file layout moved to `docs/`, with a trimmed summary + links left in place

## [0.10.0] - 2026-08-09

### Added

- Add WinGet publishing step to release workflow
- `cloudflared-version` / `cfzt-version` action inputs to pin exact releases instead of always tracking `latest`/`main` — leaving them unset keeps the previous behavior, with a workflow warning nudging toward pinning for reproducible CI
- `install.sh` accepts a `ZT_VERSION` env var to install a specific release instead of resolving `latest`

### Fixed

- **Windows:** `zt down`/`zt watchdog disable` now end the Task Scheduler task before deleting it, instead of just deleting its registration. `schtasks /delete` never stopped an already-running instance of the task's action — since that action is `zt.exe internal run` (as of v0.8.3), the leftover process was a literal "zt" process that outlived `zt down`, indistinguishable from any other running zt and enough to make `scoop update zt` refuse to update with "still running" even after every tunnel had genuinely been torn down. If you're hitting this now: `taskkill /F /IM zt.exe` (or End Task in Task Manager) once to clear the orphan, then `scoop update zt` as usual — this fix only prevents new orphans, it doesn't reach back and clean up ones from before it
- **Windows:** `zt down`/`zt watchdog disable` now also wait for the task to actually leave the Running state after ending it, before deleting its registration — `schtasks /end` only requests termination and returns immediately, so a slow-to-die cloudflared could still be alive and holding its port right after teardown reported success. If it's still running once the (short) wait is up, teardown proceeds anyway and prints a warning rather than blocking indefinitely on a possibly-stuck process
- **GitHub Action:** the "down" step matched deployments to mark inactive by `environment` alone. Since `environment` defaults to the tunnel `name` but can be overridden, several previews sharing the same explicit `environment:` value would have one preview's teardown mark every other preview's deployment inactive too. Deployments are now additionally tagged with `task: "cfzt:<name>"` at creation and matched on that, which is unique per preview regardless of `environment`
- **GitHub Action:** if recording the GitHub Deployment failed after `zt up` had already succeeded, the tunnel was left running with nothing in the workflow able to find or tear it down later. The action now runs `zt down --remote` as a cleanup step in that specific case
- **GitHub Action:** `Validate inputs` now checks for `gh`/`jq` on `PATH` up front with an actionable error, instead of failing several steps later with a bare "command not found" on a runner that doesn't have them preinstalled
- **GitHub Action:** the Cloudflare API token written to `~/.zt-config.json` for the duration of the run is now removed at the end of it — only if the run itself created the file, never a config a self-hosted runner already had
- `zt up --docker` picked a container's exposed port by iterating a Go map, whose order is randomized — a container publishing more than one port (`-p 8080:80 -p 8443:443`) could get a different port picked on different runs of the identical command. Port selection is now deterministic: the lowest TCP container port is always chosen

## [0.9.0] - 2026-08-08

### Added

- `casablanque-code/cfzt` is now also a composite GitHub Action wrapping `zt up`/`zt down` for PR preview environments, with a GitHub Deployment + Deployment Status reflecting the tunnel in the PR's own UI — see the new "GitHub Action" section in the README for usage
- `zt down <name> --remote` resolves the tunnel directly from Cloudflare by name instead of requiring it in local state — for tearing down from a different machine than the one `zt up` ran on. This is what makes the action's `down` mode work at all, since "up" and "down" are separate jobs on unrelated ephemeral runners with no state to share. Not automatic on a plain `zt down`: Cloudflare Tunnels have no "created by zt" marker the way a zt-created DNS record does, so resolving purely by name means whatever tunnel has that name gets deleted — opt-in only

## [0.8.3] - 2026-08-06

### Fixed

- **Windows:** Task Scheduler actions no longer go through `cmd.exe /c ... >> log 2>&1`. That wrapper meant every argument (cloudflared's path, the config path, the log path) was parsed three times before the real process saw it — Task Scheduler's XML, cmd.exe's own argument parsing, then cmd.exe re-parsing its `/c` string a second time — and was the root cause of tunnels that ran fine manually but failed under Task Scheduler with `Last Result: 1`. `zt up`/`zt watchdog enable` now register the task to call `zt.exe` directly via a new hidden `zt internal run` entrypoint, which opens the log file itself and execs the real command with real argv — one quoting layer instead of three
- **Windows:** `zt up` and `zt watchdog enable` no longer report success the instant `schtasks /run` accepts the request — they now poll briefly (up to ~800ms) to confirm the task is actually still running, and fail with the tail of its log if it exited immediately. Previously a cloudflared that crashed on startup (bad config, port already bound, etc.) still looked like a fully successful `zt up`, showing as inactive only on a later `zt status`/`zt doctor`
- **Windows:** `zt restart` gets the same immediate-crash detection as `zt up`
- **Windows:** fixed a regression from the cmd.exe removal above — `zt internal run` (and therefore every tunnel/watchdog task) was popping a visible console window on the desktop, and closing it killed cloudflared along with it. zt.exe now hides its own console window on startup when running as `internal run`, and runs the child in its own process group so it no longer shares fate with zt's console
- `zt list`'s "no tunnels" hint and the root `zt --help` text were still showing bare `zt up <name> <port>` — stale since v0.8.0 made `--allow`/`--public` required. Both now show a working example
- `zt --help` now groups commands (tunnel commands / manifest commands / system) instead of one flat alphabetical list, and the root help text no longer duplicates a hand-maintained command list that had drifted out of sync with the real command set

## [0.8.2] - 2026-08-05

### Fixed

- **Windows:** Task Scheduler actions no longer go through `cmd.exe /c ... >> log 2>&1`. That wrapper meant every argument (cloudflared's path, the config path, the log path) was parsed three times before the real process saw it — Task Scheduler's XML, cmd.exe's own argument parsing, then cmd.exe re-parsing its `/c` string a second time — and was the root cause of tunnels that ran fine manually but failed under Task Scheduler with `Last Result: 1`. `zt up`/`zt watchdog enable` now register the task to call `zt.exe` directly via a new hidden `zt internal run` entrypoint, which opens the log file itself and execs the real command with real argv — one quoting layer instead of three
- **Windows:** `zt up` and `zt watchdog enable` no longer report success the instant `schtasks /run` accepts the request — they now poll briefly (up to ~800ms) to confirm the task is actually still running, and fail with the tail of its log if it exited immediately. Previously a cloudflared that crashed on startup (bad config, port already bound, etc.) still looked like a fully successful `zt up`, showing as inactive only on a later `zt status`/`zt doctor`
- **Windows:** `zt restart` gets the same immediate-crash detection as `zt up`
- **Windows:** the previous fix introduced a regression — `zt internal run` (and therefore every tunnel/watchdog task) popped a visible console window on the desktop, and closing it killed cloudflared along with it. zt.exe now hides its own console window on startup when running as `internal run`, and runs the child in its own process group so it no longer shares fate with zt's console
- `zt list`'s "no tunnels" hint and the root `zt --help` text were still showing bare `zt up <name> <port>` — stale since v0.8.0 made `--allow`/`--public` required. Both now show a working example
- `zt --help` now groups commands (tunnel commands / manifest commands / system) instead of one flat alphabetical list, and the root help text no longer duplicates a hand-maintained command list that had drifted out of sync with the real command set

## [0.8.1] - 2026-08-05

### Fixed

- `zt doctor` and `zt restart` recovery hints ("run: zt down X && zt up X ...") now reconstruct the tunnel's actual `--allow`/`--public`/`--tcp` flags instead of printing a bare `zt up <name> <port>` — which failed outright after v0.8.0 made an access-policy flag required
- **Windows:** `zt up`'s scheduled task now actually starts immediately instead of only being registered to start at the next logon — previously `zt up` reported success (DNS, ingress, service all created) while cloudflared never actually launched, showing as an inactive tunnel in `zt status`/`zt doctor` right after a clean `zt up`
- Shell tab-completion for `zt down`/`status`/`restart`/`logs` was silently broken — argument reordering meant to support `zt <name> down` was corrupting cobra's internal completion protocol

### Added

- `zt list` and `zt status` now show each tunnel's protocol and access policy (public vs. Zero Trust with email count)

## [0.8.0] - 2026-08-04

### Changed

- **Breaking:** `zt up` (and `zt apply`) now requires either `--allow <email>` or `--public` — omitting both used to still create a Zero Trust Access application, but with an empty `--allow` list it got a bypass policy with an `Everyone` include rule, which disables Access enforcement entirely. In practice `zt up grafana 3000` and `zt up grafana 3000 --public` granted the same unauthenticated public access, contradicting the "Zero Trust" pitch. Any script or CI job calling `zt up` without an access flag needs one added
- `zt up`'s DNS upsert no longer deletes *any* pre-existing record with the target hostname — only a record that looks like zt's own (a CNAME to `*.cfargotunnel.com`, e.g. a stale tunnel from a previous run) is replaced automatically. A real A/AAAA record or a CNAME to something else is left alone and `zt up` refuses with an explanation; pass the new `--force` flag (also available as `force:` in `zt.yaml` manifests) to replace it anyway
- `--public` now prints a visible warning naming the hostname instead of a plain checkmark, so it's harder to miss scrolling past — chosen over a blocking confirmation prompt specifically so scripted/CI use (e.g. PR preview environments) isn't interrupted

### Added

- `zt version` (and `zt --version`) now checks GitHub for a newer release and prints an upgrade hint if one's available; `zt doctor` surfaces the same check as an informational line. Best-effort and silent on any failure (offline, GitHub unreachable) — opt out entirely with `ZT_NO_UPDATE_CHECK=1`

### Fixed

- `doctor` no longer claims "systemd linger enabled" on macOS/Windows — that check is hardcoded to always pass there (LaunchAgents and logon-trigger scheduled tasks already auto-start, there's no separate linger toggle), so the message was simply false on those platforms. Now reports what's actually being confirmed ("systemd"/"launchd"/"scheduled task" configured to auto-start at login), via the same `service.ManagerName()` used elsewhere
- `restart` printed a hardcoded `systemctl --user restart zt-<name>` on every platform, including Windows — missed when the same class of bug was fixed elsewhere. Now uses `service.ManagerName()` like `up`/`down`/`doctor`/`ls` already do

### Added

- Homebrew tap (self-hosted in this repo's `Formula/` folder, macOS + Linux) — `brew tap casablanque-code/cfzt https://github.com/casablanque-code/cfzt && brew install zt`. Auto-updates on every release via a new release.yml step that regenerates the whole formula (not a partial field patch — the Scoop bucket already broke once from exactly that)

### Internal

- Significantly expanded test coverage across `cmd/zt` and `internal/cloudflared` (previously thin or untested: config generation, version parsing, and most CLI command error paths). Caught one real bug along the way: `cloudflared.Version.TooOld()` reported `true` for an unparseable `--version` output instead of the intended "don't block on what we don't recognize" — fixed
- Cloudflare API client is now injected via a swappable `newCFClient` var instead of called directly, enabling the above test coverage without hitting the real API

## [0.7.2] - 2026-08-02

### Fixed

- Windows: schtasks error output is emitted in the console's active codepage (e.g. CP866 on ru-RU systems) and was being captured/printed assuming UTF-8, turning every schtasks failure into mojibake. Console is now switched to UTF-8 (65001) once per process before any schtasks call
- Windows: `Install()`/`Uninstall()`/`InstallWatchdog()`/`UninstallWatchdog()` now explain the most common real-world cause of "Access Denied" — a `zt-*` task previously created or touched from an elevated (Administrator) prompt can't be overwritten from a normal session, even with `/f` — and give the exact command to clear it
- Windows: `IsActive`/`WatchdogIsActive` parsed schtasks' localized `Status:`/`Running` text output, which silently always returned `false` on non-English Windows (e.g. ru-RU). Now uses PowerShell's `Get-ScheduledTask`, whose `State` is a culture-invariant enum

## [0.7.1] - 2026-08-02

### Added

- Windows: Scoop bucket (self-hosted in this repo's `bucket/` folder) — `scoop bucket add cfzt https://github.com/casablanque-code/cfzt && scoop install cfzt/zt`

### Fixed

- Windows: `zt up` and `zt watchdog enable` failed to install a real service ("ERROR: The task XML is malformed") whenever the generated command contained an XML special character — notably the `&` in the cmd.exe log-redirection wrapper's `2>&1`, which broke on every install. Silently fell back to PID mode (no auto-restart) instead of erroring loudly. Task XML values are now properly XML-escaped
- CI: Scoop bucket auto-update never actually ran — `release: types: [published]` doesn't fire for releases created with the default `GITHUB_TOKEN`, so `scoop update zt` kept reporting 0.7.0 as latest after this release shipped. Bucket bump now happens inline in the release workflow itself instead of a separate triggered workflow

## [0.7.0] - 2026-08-01

### Added

- Windows: `zt up` now installs a real Task Scheduler service (`zt-<name>`, logon trigger, restart-on-failure) instead of only running cloudflared as a directly tracked process — gives Windows the same auto-start/auto-restart guarantee systemd/launchd provide on Linux/macOS, with no admin rights or stored credentials required
- Windows: `zt watchdog` is now available, running as its own `zt-watchdog` scheduled task
- CI: `windows-latest` added to the build/vet/test matrix (previously windows-tagged files were only cross-compiled at release time, with no tests run)

### Fixed

- `zt up`/`down`/`doctor`/`ls` no longer hardcode "systemd"/`.service` in their output — these strings were unreachable on Windows before this release (`IsInstalled` always returned `false` there) but would have been actively misleading once Task Scheduler support landed

## [0.6.1] - 2026-07-18

### Added

- Test coverage: `internal/cloudflare` (0% → 79.9%), `internal/docker` (0% → 86.2%), `config` (0% → 79.3%), `internal/state/store.go` (0% → 84.8%), plus targeted tests for `cmd/zt`'s pure/orchestration logic (`tunnelStatus`, `protocolLabel`, `protocolForExport`, `runExport`, `runApply`'s diff logic, `resolveApplyPort`)

### Fixed

- `zt apply`: a Docker-flagged service with an explicit `port:` in the manifest no longer fails when Docker itself is unreachable — `resolveApplyPort()` now checks the explicit override before querying Docker, not after (previously it queried Docker unconditionally and only used the override if the query happened to succeed)
- `internal/cloudflare`: `UpsertCNAME`/`UpsertAccessApp` upsert semantics (delete-existing-before-create) now covered by tests, catching any future regression in the `zt up` rollback path
- README: removed dead `[#anchor](#anchor)` link stubs left over from earlier patches; Windows install instructions consolidated into `## Install` (were previously duplicated/split from `## Windows support`)

### Changed

- `internal/cloudflare.Client` gained an overridable `baseURL` (via `NewClientForTesting`) and `internal/docker`'s package-level `httpClient` is now swappable in tests — both internal-only, no change to public API

## [0.6.0] - 2026-07-17

### Added

- Windows: core tunnel lifecycle now works — `zt up`/`down`/`restart`/`status`/`logs`/`doctor` run cloudflared in PID-tracked mode (the same fallback mode Linux/macOS use without systemd/launchd). `zt up` reports `(no auto-restart)` explicitly when running this way.
- README: "Windows support" section documenting what works and what doesn't yet (no persistent service, no `zt watchdog` on Windows)
- README: "Installing `zt` on Windows" — PowerShell installer that downloads the latest release, verifies its sha256, and adds it to user `PATH` via `[Environment]::SetEnvironmentVariable` (not `setx`, which silently truncates PATH past 1024 characters)

### Fixed

- `cloudflared --version` detection now works on Windows (`zt doctor` previously always reported `unknown` — the Windows codepath was a hardcoded stub even though the underlying logic was already platform-agnostic)
- `.gitignore`: `dist/` build artifacts and `*.patch` files no longer show up as untracked/dirty in `git status`

### Changed

- `golang.org/x/sys` moved from an indirect to a direct dependency in `go.mod` (now imported directly for Windows process management)

## [0.5.1] - 2026-07-14

### Added

- `zt doctor`: platform-aware `cloudflared` upgrade hint — detects Homebrew (macOS) or apt (Linux, via `dpkg -S`) ownership of the `cloudflared` binary and prints the matching upgrade command (`brew upgrade cloudflared` / `sudo apt update && sudo apt install --only-upgrade cloudflared`) instead of a bare downloads link
- README: new "Installing cloudflared" section — per-platform install/upgrade table (Homebrew, apt, generic binary, Windows), replacing the single generic link in Prerequisites

### Docs

- Noted that Windows support in `zt` itself is currently partial (see `internal/service` / `internal/cloudflared` `_windows.go` stubs — process lifecycle management not yet implemented)

## [0.5.0] - 2026-07-14

### Added

- `zt status <name>` now shows the live-negotiated protocol for `auto` tunnels (`auto (http2)` / `auto (quic)`) instead of a static `auto`, via `cloudflared.DetectEffectiveProtocol()` tailing `cloudflared.log` for connection registration and startup connectivity pre-check markers
- `--version` flag
- `zt <name> status` / `logs` / `restart` / `down` now work identically to `zt status` / `logs` / `restart` / `down` `<name>`
- `zt completion bash|zsh|fish|powershell` documented in README (was already available via Cobra, just undocumented)
- README: troubleshooting entry for `502 Bad Gateway` despite an apparently healthy QUIC connection (UDP MTU/fragmentation dropping real traffic while the pre-check's small test connection still passes) — fix is `--tcp`, not automatically detectable
- README: documented that there is no `zt update` — reinstalling via `install.sh`/`go install` overwrites the binary in place

### Fixed

- `--version` / `-v` were silently non-functional: the Makefile's `-ldflags -X main.version=...` targeted a `main.version` symbol that didn't exist in the code, so the linker dropped it without error
- root `--help` didn't surface `--tcp` or shell completion, even though both existed — only visible via `zt up --help` / by knowing to look

### Tests

- 4 tests for `DetectEffectiveProtocol`, including a fixture mirroring a real degraded-QUIC pre-check log
- 8 tests for the name-first argument reordering (`zt <name> status`, flag passthrough, aliases, no-ops)

## [0.4.0] - 2026-06-20

### Added
- `zt watchdog enable` / `disable` / `status` — background service that automatically recovers tunnels stuck on the HTTP/2 fallback after a transient UDP/QUIC blip (cloudflared never retries QUIC on its own — see [cloudflare/cloudflared#1534](https://github.com/cloudflare/cloudflared/issues/1534))
- `internal/watchdog` package — incremental log tailing, exponential backoff per tunnel (10min → 60min cap), separate runtime state file (`~/.zt-watchdog-state.json`) to avoid contending with interactive commands
- `cloudflared.LogPath()` helper
- `zt doctor` now reports watchdog status
- 14 tests covering incremental log scanning, backoff math, and runtime state corruption recovery

### Changed
- `restartTunnel(name)` extracted from `zt restart` so the watchdog reuses identical restart logic
- Only tunnels with `protocol: auto` (the default) are watched — tunnels pinned via `--protocol`/`--tcp` are left untouched, as that's a deliberate user choice

## [0.3.0] - 2026-06-19

### Added
- `zt export` — snapshot all managed tunnels to a portable `zt.yaml` manifest (credentials and tunnel IDs excluded, safe to commit to git)
- `zt apply <file>` — apply a `zt.yaml` manifest on any machine; diffs against local state and creates only what is missing, never deletes automatically
- `internal/manifest` package with `Manifest`/`ServiceSpec` types, `Load` (with validation) and `Save`
- State now persists tunnel intent: `Public`, `AllowEmails`, `DockerDetect` fields added to `Tunnel` — backward compatible with existing state files
- Tests for `internal/manifest` (roundtrip, validation, edge cases) and `internal/state` (new fields, backward compat, omitempty)
- `gopkg.in/yaml.v3` as direct dependency

### Fixed
- Missing `//go:build !windows` tag in `runner.go` caused redeclared symbol errors when cross-compiling in CI
- `go.sum` now contains correct `golang.org/x/sys` transitive checksums — `go build` works without `go mod tidy`

### Changed
- `createTunnel(tunnelOpts)` extracted from `runUp` so `apply` reuses the exact same creation logic without duplication

## [0.2.2] - 2026-06-12

### Added
- `zt doctor`: check linger status for systemd user services (warns if linger is disabled — tunnels won't survive logout)
- Tunnel status check endpoint in `zt status`
- CI badge in README

### Fixed
- Linter issues; golangci-lint config inlined into workflow

## [0.2.1] - 2026-06-06

### Added
- SHA-256 checksum verification in `install.sh`
- `zt doctor` — health check utility: validates cloudflared installation, API token, DNS resolution, tunnel and service status
- API token validation on `zt init`

## [0.2.0] - 2026-05-28

### Added
- Tunnels now run as systemd user services (Linux) / LaunchAgents (macOS) — survive reboots automatically
- `--tcp` flag to force HTTP/2 protocol (for ISPs blocking QUIC/UDP)
- `--protocol` flag: `auto` (default), `quic`, `http2`
- `zt logs` command with `-n <lines>` and `-f` (follow) flags, colorized output
- `--docker` flag — auto-detect port from a running Docker container by name
- cloudflared minimum version check (2023.x) on `zt up`

## [0.1.3] - 2026-05-26

### Fixed
- Build tags for Unix/Windows syscall compatibility (`runner.go` / `runner_windows.go`)

[0.3.0]: https://github.com/casablanque-code/cfzt/compare/v0.2.2...v0.3.0
[0.2.2]: https://github.com/casablanque-code/cfzt/compare/v0.2.1...v0.2.2
[0.2.1]: https://github.com/casablanque-code/cfzt/compare/v0.2.0...v0.2.1
[0.2.0]: https://github.com/casablanque-code/cfzt/compare/v0.1.3...v0.2.0
[0.1.3]: https://github.com/casablanque-code/cfzt/compare/v0.1.2...v0.1.3