setup
Configure the testing plugin's can't-fail checks for this repository. check prints the resolved testing config cascade, which test-lint rules the repo's lint config turns on per language (a missing one is a finding), an optional instruction line to paste, and any test glob the shipped test-weaken hook skips; apply writes your answers to docs/conventions/testing.yaml (or to the docs/conventions/testing.md block or .claude/testing.yaml when that is the team file in use). Use when: 'set up testing', 'configure the test scan', 'exclude these tests from the audit', 'add a test glob', 'which test lint rules are missing', 'testing setup'. Re-runnable.
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 30
- Forks
- 7
01 前言 Front Matter
- description
- Configure the testing plugin's can't-fail checks for this repository. check prints the resolved testing config cascade, which test-lint rules the repo's lint config turns on per language (a missing one is a finding), an optional instruction line to paste, and any test glob the shipped test-weaken hook skips; apply writes your answers to docs/conventions/testing.yaml (or to the docs/conventions/testing.md block or .claude/testing.yaml when that is the team file in use). Use when: 'set up testing', 'configure the test scan', 'exclude these tests from the audit', 'add a test glob', 'which test lint rules are missing', 'testing setup'. Re-runnable.
- argument-hint
- [check|apply]
- user-invocable
- true
- disable-model-invocation
- true
02 资源文件 2
- 脚本scripts/setup.sh16.2 KB
- 脚本scripts/setup.test.sh17.1 KB
03 SKILL.md 正文
Purpose
/testing:audit and the opt-in test-weaken hook find tests by the shipped adapters' filename globs
and judge them by the shipped rule levels. This skill shows what a repository changed about that
through the testing config, and writes it. check changes nothing; apply runs check,
asks, writes, and runs check again. No argument means check.
It never edits CLAUDE.md or AGENTS.md. The instruction line is printed for you to paste.
The config
The testing config resolves across the config-cascade layers, in order: ~/.claude/testing.yaml,
the team layer and <root>/.claude/testing.local.yaml, where <root> is the scanned file's git
toplevel, else ${CLAUDE_PROJECT_DIR}. The team layer is <root>/docs/conventions/testing.yaml.
Without it, the team layer is the yaml config block in <root>/docs/conventions/testing.md
(read for one more release), else <root>/.claude/testing.yaml; the first present wins and the
resolver prints a warning naming each one it ignores. Lists concatenate across layers; a later
layer's scalar overrides. The format is the adapters' YAML subset in every file; a glob that starts
with * must be single-quoted. docs/conventions/testing.yaml also holds /testing:run-e2e's
e2e_driver, reuse_running_instance and browser_driver, and /testing:write's
tdd_refactor_step and first_test, which the scan ignores.
The keys:
adapters:
disable: [py-unittest] # wins over enable; a non-empty enable list is an allowlist
paths:
exclude: ['legacy/**'] # relative to the repo root; ** crosses /, * does not
include: ['build/**/*.test.ts'] # reaches pruned folders; an adapter must still claim the file
adapter_dirs: [tools/test-adapters] # consumer adapters, same schema as the shipped ones
extend:
js-vitest:
files: ['*.it.ts'] # appended to that adapter's list field
rules:
rule-weak-oracle: off # off drops, warn reports without gating, error gates --check
test-weaken-block: error # test-weaken hook: deny an added skip or a removed test
The opt-in test-weaken hook runs before a Write or Edit to a test file and names what it removes:
test blocks, assertions, a changed expected value, or an added skip. By default it only tells the
agent. With test-weaken-block: error, an added skip or a removed test block is
denied until the edit carries a test-change: <reason> comment; removed assertions and changed
expected values are never denied. The agent can write that marker itself: it makes the reason
visible to reviewers, it does not prove the reason.
Removals (adapters.disable, paths.exclude, rules ... off) apply inside the scanner, so the
audit goes quiet with no plugin change; an excluded path or a disabled adapter silences
test-weaken too. Additions reach the audit at once; test-weaken covers only the shipped globs.
The test judge
The judge that asks where each changed test's expected value came from runs on demand, as
/testing:audit --judge, and in /source-control:pull-request prep. It has no plugin option and
no hook.
check (read-only)
Run the script and show its output as it prints:
bash "${CLAUDE_PLUGIN_ROOT}/skills/setup/scripts/setup.sh" check
It exits 0 with no finding, 1 when a test-lint rule is missing, 2 when a layer does not resolve (the file and line are on stderr). Its four sections:
- config: the resolved records, or the note that no layer exists.
- lint: one row per rule for each language the repository's tracked test files use. The script
reads the lint config as text: a rule named there, or its plugin referenced beside a recommended
config, reads as present. Treat a
PRESENTrow from the recommended-config path as likely rather than proven, and say so. AFINDINGis a gap to report, not a neutral absence. The rules:- JS/TS:
jest/valid-expect(eslint-plugin-jest),vitest/valid-expect(@vitest/eslint-plugin),playwright/missing-playwright-awaitandplaywright/no-focused-test(eslint-plugin-playwright). Which config each sits in is its rule page's to state: jest/valid-expect, vitest/valid-expect, playwright/missing-playwright-await, playwright/no-focused-test. As of 2026-11-10; recheck when a plugin release note moves one of these rules out of its recommended config or renames it. - C#:
xUnit2021(xunit.analyzers, which the check counts as present through anxunitorxunit.v3reference) andNUnit2009(NUnit.Analyzers, which the check requires as its own reference). Packaging: xUnit2021 and NUnit2009. As of 2026-11-10; recheck when an xunit or NUnit release note changes how the analyzers are packaged. - Python: ruff
PLR0124,PT011andF631. The check countsPLR0124andF631as on with noselect,PT011as off, and aselectas replacing the defaults with prefix matching (setup.shruff_rule). Pointer: ruff settings § lint.select and the rule pages for PLR0124, PT011 and F631. As of 2026-11-10; recheck when a ruff release note changes the default rule set. - Bash, PowerShell and Go: no maintained rule.
- Any other detected language: named with
setup has no lint check for this language, not a finding.
- JS/TS:
- instruction: the optional line, naming the one file Claude Code loads at the repository
root:
CLAUDE.md, else.claude/CLAUDE.md, elseCLAUDE.local.md, elseAGENTS.md, elseCLAUDE.mdin${CLAUDE_CONFIG_DIR:-~/.claude}(the script's order). Which files load, and whenAGENTS.mdis read: memory § When Claude Code reads AGENTS.md. Offer it; do not paste it anywhere. - hook-entry: a
none:line saying test-quality findings come from/testing:audit, not a per-edit hook, then, when a consumer glob matches no shippedtest-weakenrow, a line naming those globs: the audit scans them,test-weakenskips them. No settings entry is printed.
Then probe the hook launcher, which the script does not: run command -v node via Bash and report
node as a FAIL row when it is absent. Every hook row launches through node hooks/exec-bash.mjs,
so a missing node is a hook launch error, not a skip notice, and a hook cannot report its own
missing launcher.
apply
-
Run
checkand summarize the effective config before proposing a change. Nothing already configured is dropped without the user confirming. -
Ask only what the repository cannot answer: folders or files to leave out, test files the shipped globs miss (and which adapter reads them), adapters to turn off, rule levels.
-
Write the team layer with the answers. The script writes the team file in use:
docs/conventions/testing.yamlwhole (keeping itse2e_driver,reuse_running_instance,browser_driver,tdd_refactor_stepandfirst_testlines), else the body of theyaml configblock indocs/conventions/testing.md(the rest of that file stays), else.claude/testing.yamlwhole; with none, it createsdocs/conventions/testing.yaml. It keeps the result only when it resolves and writes nothing else; pass every value the team layer should hold, the existing ones included:bash "${CLAUDE_PLUGIN_ROOT}/skills/setup/scripts/setup.sh" apply \ --exclude 'legacy/**' --extend js-vitest.files='*.it.ts' --rule weak-oracle=warnFlags:
--include,--exclude,--enable,--disable,--adapter-dir,--extend <id>.<field>=<value>,--rule <rule>=off|warn|error, each repeatable. -
Run
checkagain. Offergit addfor the fileapplyprinted and a commit, and run each only when the user accepts. Personal overrides go in.claude/testing.local.yaml, which should be gitignored.checkalso prints the optional line to paste intoCLAUDE.mdorAGENTS.mdthat points atdocs/conventions/testing.mdwhen that file exists.
Next
/testing:audit Runs the scan with the config this skill wrote.
What this skill does NOT do
- Edit
CLAUDE.md,AGENTS.md,.claude/settings.json, or any lint config. It prints what to add. - Install a lint plugin or analyzer.
- Run the audit. That is
/testing:audit.
04 同仓库更多 Skills
track
Backlog reads through the bound tracker: search open and closed items, view one item with its comments and state history. Use when another skill searches or reads tracker items.
reviewer-media
Reviewer media for one pull request: decides whether it needs a narrated walkthrough, builds the storyboard from /review:explain-change's digest, has /explainer-video:produce render it and hands the manifest to /source-control:pull-request media. Use when asked for 'reviewer media', 'a walkthrough video for this PR', 'a produced video of this change', or when pull-request ready or monitor hands a PR over.
route
Model and effort per agent role (orchestrator, worker, verifier, retrieval) for a multi-agent run, resolved through the bundled defaults and the user, team and personal override config layers into fenced JSON with single-agent and fan-out variants and the layer behind each value. Use when asked 'which model for the subagents', 'route the roles' or 'model routing', or before a skill passes a role's model on a spawn.
route
Model and effort per agent role (orchestrator, worker, verifier, retrieval) for a multi-agent run, resolved through the bundled defaults and the user, team and personal override config layers into fenced JSON with single-agent and fan-out variants and the layer behind each value. Use when asked 'which model for the subagents', 'route the roles' or 'model routing', or before a skill passes a role's model on a spawn.
principles
Module-design vocabulary for architecture findings, and the checks built on it. Use when: 'what is a deep module', 'what is a seam', 'deep vs shallow module', 'what is the deletion test', 'leverage and locality', 'module design vocabulary', 'which word do we use for X in architecture findings'. Scanning a codebase for friction is /architecture:improve.
write
Structured five-field bug report (title, steps to reproduce, expected vs actual, justified severity, suggested fix location) from an informal description, read-only. Use when the user names a defect they observed, or asks for one written up ('report a bug', 'file a bug', 'write this up as a bug').