repo-release-tools keeps release policy boring in the best possible way.
Use it from GitHub Marketplace when you want CI to validate branch names,
commit subjects, and changelog policy. Install it from PyPI when you want a
local CLI, hook integration, version bumps, and release-branch automation.
Use the Python package for local workflow automation
Choose the package if you want the developer-side tools: branch helpers,
version bumps, config inspection, pre-commit hooks, and release automation.
The Python package is published on PyPI
and has a CI counterpart in the GitHub Action guide.
For basic versioning, bump and ci-version can run without [tool.rrt] by
auto-detecting repo-root pyproject.toml, package.json, Cargo.toml,
.rrt.toml, or .config/rrt.toml.
If multiple version files are found, they are updated together. Explicit config
is for the nice extras: grouped releases, changelog paths, release branches,
lock commands, generated files, and custom patterns.
Version targets also support common language/project files such as Python
(pep621, python_version), Node/JS/TS (package_json), Go (go_version),
Rust (cargo_toml), and .NET (csproj) so multi-language repositories can
keep their release versions aligned. A mcp_server_json target keeps an MCP
Registry server.json (top-level version, each package's version, and
any oci package's image tag) in sync alongside a project's primary target.
Changelog workflows
Pick the style that matches how your repository lands changes.
incremental (default) — for teams that maintain changelog entries during development.
rrt-update-unreleased and rrt-changelog hooks stay active.
The GitHub Action resolves changelog-strategy: auto to per-commit.
rrt bump defaults to auto.
squash — for repositories that squash many commits into one PR merge.
Changelog write and check hooks skip enforcement.
The GitHub Action resolves changelog-strategy: auto to release-only.
Native config is also supported in package.json ("rrt": { ... }) and
Cargo.toml ([package.metadata.rrt] / [workspace.metadata.rrt]). Go repos
should use .rrt.toml or .config/rrt.toml.
What the project includes
rrt CLI for branches, bumps, config inspection, and Git helpers
rrt-hooks for pre-commit, lefthook, husky, and CI validation
a reusable GitHub Action in action.yml
bundled agent skills for uvx and installed-CLI workflows
docs for branch policy, hook setup, and release workflows
Get your AI agent to actually use rrt
If you use Claude Code, Copilot, Cursor, or Codex on a repository that has rrt
configured, the agent will happily reimplement what rrt already does — hand-editing a
version string in three files, inventing a branch name the pre-commit hook then rejects,
or writing a changelog entry in the wrong section. Not because it lacks the tools, but
because nothing told it these are the tools for this job.
Two things fix that. Do both.
1. Tell your agent, once, in its instruction file
Paste the block from Agent instruction snippet below into
your repository's CLAUDE.md, AGENTS.md, or .github/copilot-instructions.md. This is
the highest-leverage step by a wide margin: it is read on every session, before the agent
has formed a plan, and it costs nothing at runtime.
2. Connect the MCP server (optional, but better for anything that writes)
uv add "repo-release-tools[mcp]"
Then add .mcp.json at the repository root — Claude Code picks it up on next start:
With the server connected, most tools give the agent typed JSON instead of terminal
output it has to parse (the four lock readers and rrt_config return raw dicts instead),
commit subjects and branch names are passed as arguments instead of through shell
quoting, and mutating operations default to a dry-run preview. See the
MCP Server guide for
Claude Desktop, global install, and HTTP transport with bearer auth.
The MCP server does not cover everything. rrt docs map, rrt docs generate,
rrt docs publish, rrt docs inject, rrt tree --check, rrt toc, rrt changelog lint,
rrt changelog compare, rrt drift generate/check, rrt artifacts --check, every other
--snapshot write, and every rrt-hooks subcommand are CLI-only — a mixed session is
expected, not a fallback.
Prompt phrasings that work
These reliably route to rrt rather than to hand-editing:
"Check with rrt whether this branch name will pass the hooks before you create it."
"Use rrt to preview a minor bump — dry run — and show me every file it would touch."
"Read the Unreleased changelog with rrt before adding an entry, so you don't duplicate one."
"Run rrt doctor and tell me which hook integrations are missing."
"Before you write that commit message, validate the subject with rrt."
"What version is this repo at according to rrt?" — not "what's the version", which
invites reading a random file.
"Run rrt release check before you open the PR."
The pattern: name rrt explicitly, and name the moment ("before you create it",
"before you open the PR"). Agents route on triggers, not on capabilities.
Agent instruction snippet
## Use `rrt` for release policy
This repo uses `repo-release-tools` (`rrt`) to enforce branch naming, Conventional
Commits, changelog format, and version consistency. Do not hand-roll any of it.
Before you act, use `rrt`:
| When you are about to… | Use |
|---|---|
| create a branch | `rrt branch new <type> "<desc>"` — or validate the name first with `rrt-hooks check-branch-name --branch <candidate>` |
| write a commit message | `rrt git commit --type <type> "<description>"`, which builds and validates the subject before committing |
| change a version number anywhere | `rrt bump <level> --dry-run` first — never edit version strings by hand; pins and the changelog move with it |
| add a changelog entry | read the existing `[Unreleased]` first; it is hook-managed |
| open a PR | `rrt release check` and `rrt doctor` |
Rules:
- Every mutating `rrt` command takes `--dry-run`. Use it first, show the user the
preview, and only apply after they confirm.
- Never edit a version string by hand in more than one file — that is what `rrt bump` is for.
- Never hand-edit the `[Unreleased]` changelog section while the rrt hooks are active.
- If `rrt` is connected over MCP, prefer the `mcp__rrt__*` tools over shelling out:
typed responses for most tools, no shell quoting of commit subjects, and dry-run is
the default. Shell out for anything with no MCP tool (`rrt docs map`, `rrt docs
generate`, `rrt docs publish`, `rrt docs inject`, `rrt tree --check`, `rrt toc`,
`rrt changelog lint`, `rrt changelog compare`, `rrt drift generate`/`check`,
`rrt artifacts --check`, other `--snapshot` writes, `rrt-hooks *`).
-`rrt --help` lists every command. Check it before concluding rrt cannot do something.
repo-release-tools is released under the MIT License.
Some workflow ideas were initially inspired by
joseluisq/gitnow, but the rrt git
surface is intentionally narrower and reshaped around conventional branching,
safe commits, and release automation.
Config-driven branch and release helpers for Git repositories
The pypi package repo-release-tools receives a total of 244 weekly downloads. As such, repo-release-tools popularity was classified as not popular.
We found that repo-release-tools demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago.It has 1 open source maintainer collaborating on the project.
The compromise affects MemTensor's MemOS, an open source memory framework for large language models (LLMs) and AI agents. Both npm package @memtensor/memos-cloud-openclaw-plugin and the PyPI package MemoryOS are compromised. They drop cross-platform Go binaries that exfiltrate developer secrets.