ferrflow

FerrFlow

Universal semantic versioning for monorepos and classic repos.

Reads your Conventional Commits, works out the right version bump,
updates your version files, writes the changelog, and cuts the tagged release. Any language, any repo layout.

Documentation | Changelog | GitHub App

Why FerrFlow?

One compiled binary, no runtime to install. Native monorepo support, 16 version-file formats across every ecosystem below, and it works with whatever layout your repo already has.

ferrflow release

That is the whole release: bump, changelog, tag, GitHub release.

Supported version files

Format	File	Ecosystem	Selector
toml	Cargo.toml	Rust	package.version
toml	pyproject.toml	Python	project.version or tool.poetry.version
json	package.json	Node.js	version
json	composer.json	PHP	version
xml	pom.xml	Java / Maven	first <version> that's a direct child of the root element (skips <parent> and dependencies)
csproj	*.csproj	.NET (C#, F#)	<Version> in <PropertyGroup>
gradle	build.gradle, build.gradle.kts	Java / Kotlin	version = "…"
helm / chartyaml	Chart.yaml	Kubernetes / Helm	top-level version:
pubspecyaml	pubspec.yaml	Dart / Flutter	top-level version:
galaxyyml	galaxy.yml	Ansible collections	top-level version:
mixexs	mix.exs	Elixir	version: "…" in def project
gemspec	*.gemspec	Ruby	s.version = "…"
packageswift	Package.swift	Swift	top-level let <name>Version = "…"
cabal	*.cabal	Haskell	top-level version: field
cmake	CMakeLists.txt	C / C++	VERSION argument of project()
gomod	go.mod	Go	git tag only, no file write
txt	VERSION, VERSION.txt	Any	entire file content

Installation

Cargo

cargo install ferrflow

npm

npm install -D ferrflow

Docker

docker run --rm -v "$PWD:/repo" ghcr.io/ferrlabs/ferrflow:latest check

The image runs as a non-root user (ferrflow, uid 1000) and works on /repo. If your checkout is owned by a different uid, pass your own:

docker run --rm -u "$(id -u):$(id -g)" -v "$PWD:/repo" ghcr.io/ferrlabs/ferrflow:latest check

Pre-built binaries

Download from Releases.

Usage

# Preview what would be bumped
ferrflow check

# Adjust the plan, then get the command that reproduces it
ferrflow plan --interactive

# Run a release
ferrflow release

# Dry run
ferrflow release --dry-run

# Force a specific version (skips commit analysis)
ferrflow release --force-version 2.0.0          # single repo
ferrflow release --force-version api@3.0.0      # monorepo

# Repeatable in a monorepo, and packages can be left out entirely
ferrflow release --force-version core@3.0.0 --force-version api@2.5.0
ferrflow release --exclude web --exclude docs

# Pre-release
ferrflow release --channel beta

# Undo a release that failed partway
ferrflow rollback                    # prints the plan, changes nothing
ferrflow rollback --yes              # applies it
ferrflow rollback --yes api web      # only these packages

# Scaffold a config file
ferrflow init

# Scaffold a config file in a specific format
ferrflow init --format json5

# Use a specific config file
ferrflow check --config path/to/ferrflow.toml

# Or set via environment variable
FERRFLOW_CONFIG=path/to/ferrflow.toml ferrflow check

# Print current version
ferrflow version              # single repo
ferrflow version api          # monorepo, specific package

# Print last release tag
ferrflow tag
ferrflow tag api

# Print the dependency graph, the release order, and any cycle
ferrflow graph
ferrflow graph --json

# See what releasing a package would drag along with it
ferrflow graph --impact shared
ferrflow graph --impact shared --bump major

# Compare each package's public API against its last tag (Rust: needs cargo-semver-checks)
ferrflow api-check
ferrflow api-check --json

# JSON output (for scripting)
ferrflow version --json
ferrflow tag --json

# Shell completions
ferrflow completions bash >> ~/.bash_completion
ferrflow completions zsh  > ~/.zfunc/_ferrflow
ferrflow completions fish > ~/.config/fish/completions/ferrflow.fish

Pre-generated completion scripts are also available as ferrflow-completions.tar.gz in each GitHub release.

Configuration

FerrFlow looks for a config file at the root of your repository, in this order:

• ferrflow.json
• ferrflow.json5
• ferrflow.toml
• .ferrflow (dotfile, JSON format)

If multiple config files exist, FerrFlow exits with an error listing the conflicting files. Use --config <path> (or FERRFLOW_CONFIG env var) to specify which one to use. If no config file is found, FerrFlow auto-detects common version files in the current directory.

Run ferrflow init to scaffold a config file interactively. Use --format to skip the format prompt:

ferrflow init                  # asks which format (default: json)
ferrflow init --format json5
ferrflow init --format toml
ferrflow init --format dotfile # generates .ferrflow

Already using semantic-release? ferrflow migrate reads your existing config and generates the equivalent ferrflow.json:

ferrflow migrate                        # auto-detects .releaserc
ferrflow migrate --from semantic-release

It maps tagFormat, branches, and the common plugins (changelog, exec, github/gitlab) to their FerrFlow equivalents, and prints a summary of what mapped, what was ignored, and what needs manual review. Anything without a FerrFlow equivalent is surfaced, never guessed. JSON .releaserc is supported today; YAML .releaserc and JS release.config.js are reported as unsupported rather than mis-parsed.

JSON Schema

Add $schema to get autocompletion and validation in VS Code, WebStorm, and any JSON-aware editor:

{
  "$schema": "https://ferrflow.com/schema/ferrflow.json"
}

JSON (default)

{
  "$schema": "https://ferrflow.com/schema/ferrflow.json",
  "workspace": {
    "remote": "origin",
    "branch": "main"
  },
  "package": [
    {
      "name": "my-app",
      "path": ".",
      "changelog": "CHANGELOG.md",
      "versionedFiles": [
        { "path": "package.json", "format": "json" }
      ]
    }
  ]
}

JSON5

{
  workspace: {
    remote: "origin",
    branch: "main",
  },
  package: [
    {
      name: "my-app",
      path: ".",
      changelog: "CHANGELOG.md",
      versionedFiles: [
        { path: "package.json", format: "json" },
      ],
    },
  ],
}

TOML

[workspace]
remote = "origin"
branch = "main"

[[package]]
name = "my-app"
path = "."
changelog = "CHANGELOG.md"

[[package.versioned_files]]
path = "Cargo.toml"
format = "toml"

Monorepo

JSON

{
  "package": [
    {
      "name": "api",
      "path": "services/api",
      "changelog": "services/api/CHANGELOG.md",
      "sharedPaths": ["services/shared/"],
      "versionedFiles": [
        { "path": "services/api/Cargo.toml", "format": "toml" }
      ]
    },
    {
      "name": "frontend",
      "path": "frontend",
      "changelog": "frontend/CHANGELOG.md",
      "versionedFiles": [
        { "path": "frontend/package.json", "format": "json" }
      ]
    }
  ]
}

TOML

[[package]]
name = "api"
path = "services/api"
changelog = "services/api/CHANGELOG.md"
shared_paths = ["services/shared/"]

[[package.versioned_files]]
path = "services/api/Cargo.toml"
format = "toml"

[[package]]
name = "frontend"
path = "frontend"
changelog = "frontend/CHANGELOG.md"

[[package.versioned_files]]
path = "frontend/package.json"
format = "json"

One file per package

A large monorepo does not have to keep every package in the root config. List the per-package files under include, and each project owns its own settings:

{
  "workspace": { "versioning": "semver" },
  "include": ["services/*/ferrflow.json", "frontend/ferrflow.json"]
}

{
  "name": "api",
  "changelog": "CHANGELOG.md",
  "sharedPaths": ["../shared/"],
  "versionedFiles": [{ "path": "Cargo.toml", "format": "toml" }]
}

Paths inside an included file are relative to that file, and path defaults to its directory. The example above needs no path: services/api/ferrflow.json describes the package in services/api, and moving the directory does not require editing anything.

Included files use the same keys as a package entry, so dependsOn still refers to packages by name and works across files. They may use a different format than the root config, and they can be mixed with an inline package array while you migrate.

The following are rejected rather than silently ignored: an include pattern matching no file, two packages sharing a name, and an included file declaring workspace, include, or package.

Versioning Strategies

Each package can use its own versioning strategy. Set a default at the workspace level and override per package:

[workspace]
versioning = "semver"  # default for all packages

[[package]]
name = "api"
path = "packages/api"
# inherits semver from workspace

[[package]]
name = "site"
path = "packages/site"
versioning = "calver"  # override: date-based

Strategy	Format	Example	Description
semver	MAJOR.MINOR.PATCH	1.4.2	Default, driven by conventional commits
calver	YYYY.M.D	2025.3.28	Date-based, ignores commit types
calver-short	YY.M.D	25.3.28	Compact date-based
calver-seq	YYYY.M.SEQ	2025.3.3	Date + daily sequence counter
calver-short-seq	YY.M.SEQ	25.3.3	Compact date + sequence counter
sequential	N	42	Simple incrementing build number
zerover	0.MINOR.PATCH	0.15.2	Permanently unstable, never hits 1.0

Tag Template

By default, FerrFlow tags single-repo releases as v1.2.3 and monorepo releases as api@v1.2.3. Customize with tag_template at the workspace or package level using {name} and {version} placeholders.

[workspace]
tag_template = "v{version}"  # all packages: v1.2.3

[[package]]
name = "api"
path = "packages/api"
tag_template = "{name}/v{version}"  # override: api/v1.2.3

Layout	Default template	Example tag
Single repo	v{version}	v1.2.3
Monorepo	{name}@v{version}	api@v1.2.3
Custom	release-{version}	release-1.2.3

Pre-release Channels

Publish pre-release versions (alpha, beta, rc, dev) using the --channel flag or branch-based configuration. Pre-release versions follow the format MAJOR.MINOR.PATCH-CHANNEL.IDENTIFIER.

CLI flag

ferrflow release --channel beta       # 2.0.0-beta.1
ferrflow check --channel rc           # preview pre-release version

Branch-based configuration

Map branches to channels automatically:

{
  "workspace": {
    "branches": [
      { "name": "main", "channel": false },
      { "name": "develop", "channel": "dev", "prereleaseIdentifier": "timestamp" },
      { "name": "release/*", "channel": "rc" }
    ]
  }
}

Branch names support glob patterns. The first match wins. Wildcards match across / separators, so * matches branches like fix/global and feature/* matches feature/auth/oauth.

Identifier strategies

Strategy	Example	Description
increment	-beta.3	Auto-incrementing counter (default)
timestamp	-dev.20250402T1430	UTC timestamp
short-hash	-dev.a1b2c3d	Git short hash
timestamp-hash	-dev.20250402T1430-a1b2c3d	Timestamp + hash

Behavior

• Floating tags (e.g. v1, v1.2) are never moved by pre-release versions
• GitHub Releases are marked as pre-release
• Stable releases include all commits since the last stable tag (skipping pre-release tags)
• Hook environment includes FERRFLOW_CHANNEL and FERRFLOW_IS_PRERELEASE

Release Commit Mode

Controls how FerrFlow commits version bumps and changelog updates after a release.

[workspace]
release_commit_mode = "commit"  # default

Mode	Description
commit	Push a release commit directly to the branch
pr	Create a pull request with the release changes
none	Skip committing entirely (useful when another tool handles it)

When using pr mode, auto_merge_releases controls whether the PR is automatically merged:

[workspace]
release_commit_mode = "pr"
auto_merge_releases = true  # default

Release Commit Scope

In monorepo mode, controls whether all package bumps go into a single commit or one commit per package:

[workspace]
release_commit_scope = "grouped"  # default

Scope	Description
grouped	Single commit for all packages (e.g. chore(release): api v1.0.0, site v2.1.0)
per-package	One commit per package (e.g. chore(release): api v1.0.0, then chore(release): site v2.1.0)

Per-package commits make it easier to revert a single package bump without affecting others. This works with both commit and pr release modes.

Skip CI

By default, release commits in commit mode include [skip ci] in the message to avoid triggering a CI loop. Override with skip_ci:

[workspace]
skip_ci = false  # force CI to run on release commits

In pr mode, skip_ci defaults to false since the PR merge triggers CI naturally.

How pr mode releases

pr mode runs in two phases, so nothing is published for a release that has not been accepted.

The proposing run computes the bump, writes the version files and changelog onto the release branch, and opens or updates the pull request. No tags are created and no releases are published. Every later commit on the target branch regenerates the branch, so the open PR keeps tracking the version that would ship now.

The finalising run happens after the PR merges. FerrFlow sees a chore(release): commit on the target branch whose versions carry no tag, and tags exactly those versions before publishing the releases. The versions are read from the version files, never recomputed, so merging a release PR cannot cascade into a further bump. Squash merges and merge commits both work.

A package declared without versionedFiles has no version to read, so it is not finalised this way. Use commit mode for tag-only packages.

Rollback

A release that fails partway leaves tags pushed, forge releases created and a release commit on the branch. ferrflow rollback undoes exactly what that run did, reading the checkpoint it left behind rather than guessing from the log.

ferrflow rollback

It prints the plan and changes nothing. Add --yes to apply it. Name packages to narrow it to a subset; with none, it rolls back everything the run touched.

Three things it deliberately refuses to do.

It never deletes a tag that has moved. Each tag is recorded with the commit it pointed at, and the remote is queried at rollback time to check it still points there. One that no longer matches, or that this checkout cannot resolve on the remote at all, is reported and skipped: a tag someone else recreated in the meantime is not this run's to remove.

It stops on a package already published to a registry that cannot be unpublished. crates.io and PyPI keep every version forever, and npm refuses to republish an unpublished one, so deleting the tag would leave a version anyone can install with nothing pointing at it. The right answer there is a new patch version, and rollback says so instead of doing half the job. Docker tags, Helm charts, release assets and webhooks are replaceable, so they do not block anything.

It reverts the release commit only on a whole-run rollback with nothing blocked. That commit carries every package's version bump, so reverting it while one package stays released would silently undo that package's version too.

The revert is left uncommitted to the remote on purpose. Rollback has just deleted remote refs, and forcing a branch update on top of that is a decision worth taking with the branch in front of you.

Floating Tags

Move abbreviated tags (e.g. v1, v1.2) to always point at the latest matching release:

[workspace]
floating_tags = ["major"]  # creates/moves v1 when releasing v1.2.3

Level	Tag	Points to
major	v1	Latest v1.x.x
minor	v1.2	Latest v1.2.x

Floating tags are never moved by pre-release versions. Override per package:

[[package]]
name = "api"
path = "packages/api"
floating_tags = ["major", "minor"]

Orphaned Tag Strategy

After a rebase + force-push, existing tags may point to commits that no longer exist on the branch. orphaned_tag_strategy controls how FerrFlow handles this:

[workspace]
orphaned_tag_strategy = "warn"  # default

Strategy	Description
warn	Log a warning and skip the orphaned tag
treeHash	Attempt recovery by matching the commit's tree hash
message	Attempt recovery by matching the commit message

Recover Missed Releases

In monorepos, a package can miss a release if its files changed but FerrFlow wasn't run. Enable recover_missed_releases to compare files against the last tag instead of just the last commit:

[workspace]
recover_missed_releases = true  # default: false

Package Dependencies

In a monorepo, use depends_on to automatically patch-bump a package when one of its dependencies is released:

{
  "package": [
    { "name": "core", "path": "packages/core" },
    {
      "name": "cli",
      "path": "packages/cli",
      "depends_on": ["core"]
    }
  ]
}

When core is bumped, cli gets a patch bump even if it had no direct commits.

Hooks

Run shell commands at lifecycle points during a release. Hooks can be set at the workspace level (applies to all packages) or per package:

[workspace.hooks]
pre_bump = "echo 'about to bump'"
post_bump = "cargo check"
pre_commit = "npm run build"
pre_publish = "npm pack --dry-run"
post_publish = "notify-slack.sh"
on_failure = "abort"  # or "continue"

Hook	When
pre_bump	After bump calculation, before writing version files
post_bump	After writing version files, before changelog generation
pre_commit	After changelog generation, before git commit
pre_publish	After commit and tag, before push
post_publish	After push and release creation

If a hook exits non-zero and on_failure is abort (default), the release is cancelled. Set on_failure to continue to ignore hook failures.

Hook commands receive environment variables: FERRFLOW_PACKAGE, FERRFLOW_OLD_VERSION, FERRFLOW_NEW_VERSION, FERRFLOW_BUMP_TYPE, FERRFLOW_TAG, FERRFLOW_PACKAGE_PATH, FERRFLOW_DRY_RUN, FERRFLOW_CHANNEL, FERRFLOW_IS_PRERELEASE.

Conventional Commits

FerrFlow reads Conventional Commits to decide how far to bump. The spec is the contract between your history and your version numbers: write the prefix, and the bump, the changelog section and the tag follow from it.

Prefix	Bump
fix:, perf:, refactor:	patch
feat:	minor
feat!:, BREAKING CHANGE	major
chore:, docs:, ci:	none

The defaults are deliberately permissive, so a repo that never enforced the spec still gets sensible bumps: capitalised and slash-separated variants (Feat:, Feat/, feature:, Fix/, Perf:, Refactor/, and so on) are accepted alongside the canonical forms. If your history uses different conventions, remap them with commitFormats.

Nothing maps to major by default. A major bump comes only from a structural breaking marker (feat!:, fix(api)!:, or a BREAKING CHANGE: footer), and those are always detected whatever you configure, so a commit cannot lose its breaking status through a custom pattern.

Full reference: ferrflow.com/docs/reference/conventional-commits.

CI usage

GitLab CI

release:
  image: ghcr.io/ferrlabs/ferrflow:latest
  script:
    - ferrflow release
  rules:
    - if: '$CI_COMMIT_BRANCH == "main"'

GitHub Actions

- name: Release
  run: ferrflow release
  env:
    GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

Using the hosted bot (ferrflow[bot])

Install the FerrFlow GitHub App on your repo or org, then opt in with bot: true. Release commits, tags, and GitHub Releases are authored by ferrflow[bot] and downstream workflows triggered by those events run normally (unlike the default GITHUB_TOKEN, which suppresses them).

permissions:
  id-token: write
  contents: read

steps:
  - uses: actions/checkout@v6
    with:
      fetch-depth: 0
  - uses: FerrLabs/FerrFlow@v7
    with:
      bot: true

That is the whole job. No setup-node, no extra dependencies. FerrFlow's Rust binary handles the OIDC exchange directly, so minimal self-hosted runners work out of the box.

Three auth modes are supported: bot: true uses the hosted FerrFlow App (recommended); token: <PAT> uses a personal access token or your own GitHub App token (DIY); omitting both falls back to the workflow's GITHUB_TOKEN (simplest, but release events won't trigger downstream workflows).

On GitHub, the release commit is authored through the forge rather than by local git when bot: true is set, which is what makes it show as verified. A GitHub App cannot register a signing key, so letting GitHub author the commit is the only way its commits are signed. GitHub does that only when the request carries no custom author, committer or signature, so the commit is attributed to ferrflow[bot] and nothing else.

Everywhere else, and without bot: true, the commit is built by local git exactly as before. If you want verified commits under your own identity, set commit.gpgsign and FerrFlow will honour it, on every forge.

Annotated tags are still created and pushed by git, so they are not covered by this. Only the commit is.

License

MIT