Testing Slash Commands

/test-unit is the go-to for most projects. It reads your package.json or pyproject.toml, detects your test framework (Vitest, Jest, Pytest, Mocha, RSpec), reads an existing test file to match your style, and colocates the new test next to the source.

Cost is low (Sonnet by default) and it's the highest-frequency testing command teams install first.

It reads your dependency manifest:

• JS — package.json devDependencies for vitest, jest, mocha, or ava.
• Python — pyproject.toml or requirements-dev.txt for pytest, unittest, or nose.
• Ruby — Gemfile for rspec vs minitest.

This is why the command works across stacks without manual config.

Slash command for on-demand, per-file work — /test-unit at the moment you finish the source file.

Subagent for longer, exploratory tasks — for example, delegating "find the top 10 coverage gaps and write tests for all of them" to a test-coverage-analyst subagent lets it work in an isolated context without polluting your main conversation.

Most teams use both.

Narrow ones. A test-writing command needs:

• Read, Grep, Glob — to understand the file being tested and existing patterns.
• Write(**/*.test.*), Write(**/*.spec.*), Write(tests/**) — scoped to test paths only.

It should not have Bash access (tests aren't a place to shell out) and should not be able to write to non-test files. Narrow allowed-tools is the safety net.

Bake determinism rules into the command body. Explicitly instruct:

• No arbitrary sleeps — use waitFor with conditions.
• Reset state between tests — use before/after hooks with cleanup.
• Fix timezone and Date.now in test setup.
• Use test fixtures for auth rather than logging in during each test.

Force model: opus in the frontmatter — E2E test design benefits from the deeper reasoning.

Yes, on the common cases. It's most effective when given the failing test and recent CI history (which it reads via git log and, if configured, an MCP server for your CI provider).

Common patterns it recognizes:

• Awaited promises resolving in different orders
• Tests depending on execution order
• Non-deterministic date/random values
• Network timeouts

For deep environmental flakiness (Docker networking, GPU driver issues), you'll still need to dig manually.

Yes, and it's one of the highest-ROI hooks a team can install. The auto-test-related hook fires PostToolUse on Write/Edit events, detects which tests import the changed file, and runs just those.

If a test fails, the output goes back to Claude so it can fix on the same turn. Combined with /test-unit for authoring new tests, this pair keeps your test suite healthy without you thinking about it.

The command should detect your project's mocking approach and match it:

• Vitest → vi.mock
• Jest → jest.mock
• Network → MSW (browser + node), nock, or vcrpy
• Spies → sinon

For network mocking specifically, /mock-external-apis generates MSW handlers, nock scopes, or vcrpy cassettes based on the API calls in the code.

Four commands cover most of what a team needs:

1. /test-unit — daily unit test authoring.
2. /test-e2e — critical user flows.
3. /coverage-check — periodic gap analysis.
4. /fix-flaky-test — the inevitable CI complaints.

Add /mock-external-apis if your project has significant third-party API dependencies. All five are on this page.

Yes. Pass the framework as an argument. For example:

/test-unit UserService.ts --framework=jest

…forces Jest even if the project uses Vitest elsewhere. Useful when you're incrementally migrating or supporting multiple test runners in a monorepo. The command's argument-hint field documents this.