Changelog¶

[1.4.0] - 2026-08-25¶

🚀 Features¶

Run a repo-owned setup hook as part of install

🐛 Bug Fixes¶

Do not ask an executable question Windows cannot answer
Reach both executable-check branches without patching os.name

[1.3.1] - 2026-08-23¶

🐛 Bug Fixes¶

(ci) Keep the workspace when pushing to the paper branch
(ci) Pin tectonic's cache path instead of inferring it
(ci) Run the paper workflow when its action changes

[1.3.0] - 2026-08-23¶

⚠️ Upgrade note¶

rhiza-task paper compiles with tectonic, and no longer with a TeX distribution's driver. A machine that could build the paper before and has no tectonic will now report skipped paper tectonic not found instead of producing a PDF — a machine that previously succeeded, so this is breaking even though the gate reports it as a skip rather than a failure. Install tectonic — brew install tectonic, cargo install tectonic, or a release binary; note that Ubuntu does not package it, which is what CI here installs a pinned tarball for — or pin rhiza-task<1.3.

What the trade buys, and what it costs:

One binary instead of a distribution plus a package list. The old install asked each workflow to name the TeX packages the document happened to cite, and this repository carried two such lists — one per workflow, compiling the same document, with nothing checking that they agreed. tectonic resolves what the document cites from its own bundle, so there is no list to diverge.
The argument vector is the document and one flag. Convergence and bibtex are tectonic's own loop rather than a driver's, and there is no interaction mode to pin because tectonic never prompts. --keep-logs is the flag, kept because tectonic writes only the PDF by default and the log is what you upload when a CI compile fails.
A cold cache needs the network, where a provisioned distribution did not. The cache is per-machine and survives between runs, so it is a first-run cost; both workflows here now cache it.

rhiza-task paper-clean no longer guards on a tool, and no longer sweeps by extension. tectonic has no clean subcommand, so the task deletes the artifacts itself — which makes it the one task in the section that works on a machine that cannot build the paper at all. It removes the PDF and auxiliary files belonging to the folder's top-level .tex documents, so a committed diagram.pdf with no diagram.tex now survives a clean that would previously have been the driver's business. If you relied on paper-clean clearing every .pdf in the folder, it no longer does.

rhiza-task doctor no longer probes for GNU make, and has no optional tier left. It names uv and git, and a miss is a failure. doctor.mk required make because the whole task layer was make; this package has none, and nothing in the registry needs it. Every other binary the package reaches for — docker, gh, git-lfs, tectonic, marp — is a Guard on the task that wraps it and reports itself on that task's skipped line with an install URL, which is where an optional prerequisite belongs. Tool.required is gone from rhiza_task.tasks.doctor with it.

🚀 Features¶

Compile the paper with tectonic, and stop probing for make

🐛 Bug Fixes¶

(ci) Install tectonic from a pinned release, not from apt

📚 Documentation¶

Add a total-downloads badge, and record what it counts

[1.2.0] - 2026-08-22¶

⚠️ Upgrade note¶

The mutation task is gone. rhiza-task mutation now exits with the CLI's unknown-task error, and make mutation falls through the shim's %: catch-all to the same error. That is the intended outcome rather than a regression: the ecosystem no longer offers a mutation gate.

It is a removal rather than a port, and the reasons are worth stating because they decide what a consumer should do instead:

The task had been broken since mutmut 3 shipped. Its vector passed --paths-to-mutate, --tests-dir and the html subcommand, all three of which mutmut 3 removed, and it provisioned mutmut unpinned — so the breakage was time-triggered rather than version-triggered, and reached every consumer at every pin with no sync involved.
The replacement is not a flag rename. mutmut 3 takes source paths from a [tool.mutmut] table and offers no CLI path for them, so a port means either requiring every consumer to declare that table or writing into their pyproject.toml — and the HTML report the task relocated no longer exists, its successor being export-cicd-stats JSON.
Nothing invoked it. No all named it, and the guarded upstream workflow was unset in every repository we could see, so the port would have been paid for here and verified nowhere.

A repository that wants mutation testing runs mutmut directly: declare [tool.mutmut] source_paths = [...] and run uvx --from mutmut mutmut run. Nothing in rhiza-task reads or writes that table.

This is a minor and not a major, which is a deliberate call rather than an oversight: the generated list below marks the removal [**breaking**], and it is one — a task name that resolved now errors. It landed as a minor because the gate had not worked since mutmut 3 shipped, so no consumer had a working invocation to lose. If you pin rhiza-task and invoke mutation, this release is breaking for you regardless of its number.

🚀 Features¶

Let an empty pytest_rhiza run the checks from the project itself
[breaking] Remove the mutation task, and every trace of it

🐛 Bug Fixes¶

Leave no partial report when mutation's relocation fails

🚜 Refactor¶

Decompose Config.post_init into per-setting validators

📚 Documentation¶

Diff layers.md's shadowing example, and correct the radon claim
Correct three stale figures in CLAUDE.md's House style section
Quote only figures that cannot drift, and say which those are
Stop giving a total for uv.py's entry points, and name capture

⚙️ Miscellaneous Tasks¶

Run the testing-extras vectors against a real toolchain

[1.1.0] - 2026-08-21¶

⚠️ Upgrade note¶

docs-examples checks more than it did at v1.0.0, so a repository whose documentation passed that gate can see it go red on upgrade without having changed anything. That is the gate working, not a regression. Three changes compound:

toml and yaml fences are parsed, where they were previously counted as uncheckable. A malformed configuration example under the docs folder now fails.
README.md's toml and yaml fences are checked too. Its python and bash fences remain pytest-rhiza's, under rhiza-test, so no fence is checked twice.
A yaml checker that crashes now fails instead of reporting "parser unavailable", which passed.

To see what a repository would report before upgrading, run uvx rhiza-task@1.1.0 docs-examples against it.

This note is here because the three changes landed as fix: and refactor: commits, so the generated sections below file them where a reader does not look for a behaviour change. The convention that a gate getting stricter is a feat: was written down after the fact, in CLAUDE.md.

🚀 Features¶

(book) Add the mkdocs book, built by the existing book task
Gate the docs tree's fenced examples, and correct four drifted figures

🐛 Bug Fixes¶

(bumpversion) Relock, and carry uv.lock through the bump
(runner) Propagate the failing task's own exit status
Contain *_folder settings, record the unwired gates, fold a deferred import
(security) Stop the security gate warning about this repo's own suppressions
Reject a --root that is not an existing directory
Stop publishing local build paths in the book's reports
Close the seven findings from the quality run
Decompose Guard to A (5), and retire the orphaned Scorecard comments
Close the three findings from the quality run
Close the three regressions the quality re-run found
Check README's data fences, which nothing was checking
Bump the version pins in docs/, not just README.md
Let a release PR go green by deselecting the tag-version check

💼 Other¶

Add a pre-commit config, so fmt measures something
Raise the coverage floor to 100, and make the stronger claim it unlocks
Automate the pins, test the parsing, and write down the invariants
Bump the actions group with 3 updates
Pin this repo's typechecker rather than inheriting the default
Run both typecheckers, and fix the one finding mypy --strict adds

🚜 Refactor¶

Promote the name normaliser, and record why Python's security gate differs
Extract the fence checker into tasks/fences.py
Promote task_modules, and say where the layering rules stop

📚 Documentation¶

(readme) Add five badges and cut the prose back by a fifth
Make the examples executable, and declare the layout opt-out
Say why the four C-complexity blocks are shaped as they are
Link the book and widen the package description
(ci) Correct the comments the pre-commit config made stale
(design) Introduce jointview, so the comments citing it have an antecedent
(build) Fix the stale premise about how the opt-out reason spells its floor
Give Config.post_init's growth rule a ceiling
Add a security policy
Correct the --strict answer, which had gone stale twice
Link upstream's decision records from the design page
Add docs/paper/paper.tex describing the package
Publish the compiled paper in the mkdocs book
Correct drifted figures and split the three example gates
Describe the toml and yaml fence checks in tasks.md
Record why fences.py is the largest module in src/
Make commit types release evidence, and flag the pending minor
CLAUDE.md still said four entry points

🧪 Testing¶

Assert exit-status and outcome contracts, not only argument vectors
Assert the propagated exit code, not the collapsed one
Cover the last 37 statements, without touching src
Execute the 39 doctests, so the documented contracts are gated
Cover the two defensive except branches, retiring their pragmas
Assert the installed version matches the declared one
Stub capture too, and assert the hermeticity guarantee

⚙️ Miscellaneous Tasks¶

(bumpversion) Keep the README's rhiza-task@ pin in step
(release) Re-sync from rhiza v1.4.2, keeping the local trim
Adopt rhiza's scorecard, codeql and quality-review stubs
Test both ends of the declared dependency range
Enforce ruff.toml in a gate CI runs
Lift the book and paper workflows from rhiza
Invoke the CLI end-to-end on every matrix leg, not just ubuntu
SHA-pin the two reusable workflow calls
Add CODEOWNERS and the branch and tag ruleset definitions
Add an editorconfig
Close the four gaps /quality found, and gate the complexity ceiling
Drop the two upstream workflow delegations, and correct CLAUDE.md
Install TeX Live on the ref the book deploys, fixing the paper 404

Changelog¶

[1.0.0] - 2026-08-20¶

🚀 Features¶

(shim) Spell out the named tasks and declare them phony
(shim) Name paper and presentation too
(shim) Drop doctor from the named set
(shim) List local.mk's own targets in make help
(shim) Provision uv alongside uvx
(cli) [breaking] Remove the Makefile shim

🚜 Refactor¶

(shim) Name the goal the catch-all forwards
(shim) Make FORCE the phony mechanism, drop the named list

📚 Documentation¶

(shim) Cut the shim's comments back, and say what belongs in it

⚙️ Miscellaneous Tasks¶

Untrack the root Makefile again

[0.3.1] - 2026-08-19¶

🐛 Bug Fixes¶

(book) Provision mkdocstrings, which the book bundle's own config requires

[0.3.0] - 2026-08-19¶

🚀 Features¶

(tasks) Tasks for the five bundles .rhiza/make.d could not retire (#22). One module per bundle — github, docker, lfs, paper, presentation — covering all eighteen targets under their original names, so make view-prs keeps working through the shim's catch-all. This reverses 0.1.0's "not ported: github.mk's seven gh wrappers (use gh directly)": github is in rhiza's github-project profile, so leaving it out kept the whole .rhiza/make.d/ folder alive for the flagship profile.
(spec) Guard(tool=...), which is what require-gh and require-marp were — a target whose entire body is command -v X >/dev/null || exit 1, declared as a prerequisite of every helper. A missing tool is a skip carrying the install hint rather than the fragment's hard exit, because nothing in these five bundles is a gate; --strict gives the failure back to a caller who wants it.
(config) docker_folder, docker_image, paper_folder, presentation_file and marp_package. docker.mk's DOCKER_FOLDER was a := — not configurable at all — and PRESENTATION.md was a literal.

Deliberate differences from the fragments¶

Three, each recorded in its module docstring:

lfs-install configures the repository; it no longer downloads a binary. lfs.mk's macOS branch queried the releases API, unzipped git-lfs into .local/bin, and ran git-lfs install — but .local/bin is not on PATH afterwards and every other target invokes the bare command, so lfs-install followed by lfs-pull still failed on a machine that had none. The git lfs install at the end is the part that worked and is what survives; the binary is now reported per platform, not fetched. Consumers who relied on the Linux sudo apt-get branch need one line of their own.
Marp comes from npx --yes, not npm install -g. require-marp did not check for Marp, it installed it globally as a side effect of typing make presentation. npx keeps the convenience without mutating the machine; Marp on PATH still wins, and marp_package pins the spec.
paper no longer prefers basanos.tex — one downstream repository's filename, hard-coded in a template every consumer syncs. Now main.tex, then paper.tex, then alphabetical, which is identical behaviour for a folder holding one .tex.

Dropped with the fragments: require-gh/gh-install, the same question spelled twice because make cannot say it once; require-marp; and FORGE_TYPE, which github.mk computed at parse time and no target ever read. The six gh templates are carried over byte-identical — reproducing timeago and gh's colour handling in Python would be a worse table.

[0.2.0] - 2026-08-18¶

🚀 Features¶

(tasks) Add the Rust and Go language layers
(config) Add rhiza.toml, and read the table from Cargo.toml too
(tasks) Add coverage to all three layers

🐛 Bug Fixes¶

(config) Split a string reaching a tuple field, and stop print eating markup
(config) Let RHIZA_CHECKS name the rhiza_checks field
(shim) Bootstrap uv when it is absent
(go) Pass the coverage profile path with forward slashes

📚 Documentation¶

(changelog) Note the Rust and Go layers

🎨 Styling¶

(config) Restore the import order two merges shuffled

[0.1.2] - 2026-08-18¶

🐛 Bug Fixes¶

(config) Treat an empty setting as unset, and floor the CI OS matrix

[0.1.1] - 2026-08-18¶

[0.1.0] - unreleased¶

Initial extraction. Replaces .rhiza/rhiza.mk and .rhiza/make.d/*.mk (1023 synced lines) with a pinned CLI, on the same principle as pytest-rhiza replacing .rhiza/tests.

core + python-core gate set: install, test, typecheck, security, deps, license, docs-coverage, fmt, semgrep, rhiza-test, todos, clean, doctor, all
tests extras: benchmark, hypothesis-test, stress
book / marimo: book, serve, marimo, marimo-validate
Not ported: github.mk's seven gh wrappers (use gh directly), install-uv (it cannot be a task, since it provisions the runtime every task runs under — the shim keeps it as a file target)

Deliberate differences from the make layer¶

Four places where the port does not reproduce the recipe, each found by running the gates against this package:

security passes --ini .bandit only when that file exists. python.mk passes it unconditionally, and bandit treats a missing ini as a usage error — so a project without one got a red security gate reporting a configuration problem.
doctor treats GNU make as optional. doctor.mk requires it, which was honest while the task layer was make. Only the shim needs it now.
The shim declares local.mk: ;. An included makefile is a target make tries to remake first, and the catch-all routed that attempt to the CLI — so every invocation began with unknown task: local.mk.
shim writes to stdout, not through rich. Console output word-wraps and expands tabs; a Makefile survives neither, so rhiza-task shim > Makefile produced a broken file.