install installs ruff/pytest without the repo’s pinned dev requirements and misses dependencies that the repo actually uses (e.g. pyyaml for tests and mpremote for repl/mount and hardware test runner). Consider installing from .github/workflows/requirements.txt (keeps versions aligned with CI) and include mpremote, using python -m pip to respect the active venv.

test-driver uses -k "$(DRIVER)", but this test suite already provides a dedicated --driver option in tests/conftest.py. Using -k will match substrings in nodeids and may select the wrong set of tests (or none). Switch this to --driver $(DRIVER) to match the documented/implemented driver filter.

These Husky hook files don’t include a shebang (e.g. #!/usr/bin/env sh). Depending on platform and how Git executes the hook, a missing shebang can cause the hook to fail to run. Add the standard Husky header (and any required initialization for your Husky major version) so the hooks execute reliably.

This Husky hook file doesn’t include a shebang (e.g. #!/usr/bin/env sh). Depending on platform and how Git executes the hook, a missing shebang can cause the hook to fail to run. Add the standard Husky header (and any required initialization for your Husky major version) so the hook executes reliably.

scope-enum is configured with severity 1 (warning), which means invalid scopes will typically not fail the commit-msg hook. If the intent is to enforce the allowed scopes (as described in the PR), set the rule severity to 2 so invalid scopes are rejected.

lint-staged@16.4.0 pulls in a Node engine requirement of >=20.17 (per package-lock.json). Since this repo previously didn’t declare a Node version, developers on Node 18 LTS will hit install/runtime errors unexpectedly. Consider either pinning lint-staged to a version compatible with your supported Node range, or declaring an explicit engines.node in package.json.

Install PyYAML in dev setup target

make install currently installs only ruff and pytest, but the test suite imports yaml (for example in tests/conftest.py and tests/test_scenarios.py). On a fresh environment, following this new workflow causes make test to fail with ModuleNotFoundError: No module named 'yaml', so the advertised setup target does not actually provision the dependencies needed for testing.

Useful? React with 👍 / 👎.

Allow existing scope-only commit subjects

The new type-enum restricts the leading token to conventional types and excludes driver scopes (e.g. hts221, ism330dl), so commit messages in the repository’s established <scope>: <Description.> format are rejected by the new commit-msg hook. This blocks contributors using the documented/project-native format unless they switch to a different convention, which creates workflow breakage.

Useful? React with 👍 / 👎.

Run Husky via npm exec in prepare target

make setup invokes prepare directly from make, but husky is only installed in node_modules/.bin and that directory is not on PATH outside npm lifecycle scripts. In a normal fresh checkout this makes make setup fail at prepare with husky: No such file or directory, even after npm install; call Husky through npx --no-install husky (or npm exec husky) here so the advertised setup command works reliably.

Useful? React with 👍 / 👎.

Install Python tools with the same interpreter used for tests

The Makefile installs dependencies with bare pip but runs tests with python3 -m pytest; on systems where pip targets a different interpreter (or is missing while only pip3 exists), make install succeeds/fails inconsistently and make test can still raise missing-module errors. Installing via python3 -m pip keeps the environment aligned with the test runner.

Useful? React with 👍 / 👎.

Match per-scenario filters to pytest scenario IDs

The generated test-<scenario> targets filter with -k "$(s)", where $(s) comes from the YAML filename stem, but test IDs are built from scenario name/driver (tests/test_scenarios.py), not necessarily from the filename. For example, tests/scenarios/wsen_hids.yaml uses driver: wsen-hids and tests/scenarios/board_temperature_comparison.yaml uses name: "Temperature comparison (all sensors)", so make test-wsen_hids and make test-board_temperature_comparison can select zero tests and fail with "no tests collected". Use the same identifier source as pytest (or --driver) when generating these targets.

Useful? React with 👍 / 👎.

Select board tests by marker instead of name substring

These targets split board vs sensor suites using -k substring checks on board_, but board scenarios are identified by the board marker in the test generator (tests/test_scenarios.py marks board scenarios explicitly). Any board scenario whose ID does not contain board_ (e.g. name: "Temperature comparison (all sensors)") is omitted from test-board and incorrectly included by test-sensors, so both commands run the wrong scope. Use marker-based filtering (-m board and -m "hardware and not board") to make the target behavior correct.

Useful? React with 👍 / 👎.

Reinstall npm deps when lockfile changes

The dependency stamp for node_modules/.package-lock.json only depends on package.json, so a pulled change that updates package-lock.json without changing package.json will not trigger npm install. That leaves node_modules potentially out of sync with the repository lockfile and can cause hook/tooling version drift between contributors. Include package-lock.json as a prerequisite (or avoid this stamp pattern) so installs remain reproducible.

Useful? React with 👍 / 👎.

Dynamic per-scenario targets are generated from YAML filenames (e.g. wsen_hids.yaml -> make test-wsen_hids), but pytest IDs use scenario.get('name', scenario.get('driver', ...)). For wsen_hids.yaml/wsen_pads.yaml, driver is wsen-hids/wsen-pads, so -k "wsen_hids" won't match and the generated make targets won't run any tests. Consider generating targets from the scenario name/driver fields (parsed from YAML) or aligning filenames/name with the driver string used in test IDs (e.g., rename YAMLs or add a name: matching the filename).

In semantic-release config, debug should be a boolean, but it's currently the string "false". As written, semantic-release will treat it as truthy and may enable debug logging unexpectedly. Change it to the JSON boolean false.

The pre-commit messages are written in French and include emojis. Most repo documentation and tooling output is in English, so this can be confusing for contributors and CI logs. Consider switching these messages to English (or making the language configurable) and keeping output plain-text for maximum compatibility.

console.log rule message also contains an emoji and uses a curly apostrophe (N’as-tu). Depending on terminal/font/encoding this can render poorly in hook output. Consider using ASCII-safe punctuation and consistent language with the rest of the repo.

Branch-name validation error message is in French and includes emojis, while the repo docs are primarily English. Consider switching to an English message (or making language configurable) so the hook output is understandable for all contributors.

pip install -e ... can pick up the wrong Python environment (e.g., if pip points to a different interpreter than python3). Using python3 -m pip install -e ... (or $(PYTHON) -m pip with a configurable PYTHON var) makes the install target more reliable.

prepare runs husky with no arguments. Husky typically needs an explicit install step (e.g. setting core.hooksPath and creating .husky/_) for hooks to actually run; calling husky alone is likely a no-op or help output. Please ensure make prepare performs the actual installation step so git hooks are activated after make setup.

The documented commit format shows <type>(<scope>): ..., but the examples include docs: ... (no scope). Conventional Commits makes scope optional, and commitlint typically doesn't require it unless explicitly configured. Suggest updating the docs to indicate that (scope) is optional (and clarify whether you want to require it or not).

Workflow section lists allowed branch prefixes as feat/, fix/, docs/, tooling/, ci/, test/, style/, but the actual validation regex also allows chore/ and refactor/. Align the documentation with the enforced pattern so contributors don't get surprised by hooks rejecting/accepting names differently.

Using $(shell pwd) makes Make spawn a subshell at parse time and can behave unexpectedly if make is invoked with a different working directory. $(CURDIR) is the standard Make built-in for the current directory and is usually more portable/reliable for constructing PATH.

Mark Husky hook scripts executable

These hook files were added with mode 100644, so Git ignores them instead of executing them (.husky/pre-commit/.husky/commit-msg are treated as non-executable). In a fresh clone this means the new commit-msg and pre-commit checks are silently bypassed, so invalid commit messages and unlinted staged Python changes can still be committed even after make setup.

Useful? React with 👍 / 👎.

Fail fast when npm install fails in setup target

Because .ONESHELL is enabled globally, this multi-line recipe runs in one shell without -e; if npm install fails but node_modules/ already exists, the subsequent touch still succeeds and the target exits 0, leaving a false success stamp. That can make make install/make setup proceed with stale or missing Node tooling while reporting success, which breaks the reproducibility this target is meant to provide.

Useful? React with 👍 / 👎.