/test-advanced - Advanced Test Planning Custom Command

/test-advanced is not a built-in Claude Code command. This entry is a custom command recipe: you create a Markdown file in .claude/commands/, Claude Code exposes it as /test-advanced, and the behavior comes from the prompt you write into that file.

Use it when the simpler /generate-tests style prompt is too broad and you want Claude to plan a test matrix before writing code: current behavior, invariants, edge cases, regression cases, framework conventions, and any advanced gaps your project already has tooling to measure.

Create the command

Create .claude/commands/test-advanced.md:

---
description: Plan and draft deeper tests for a file, function, or behavior
argument-hint: [file-or-function]
---

Plan advanced test coverage for $ARGUMENTS.

First inspect the target code and nearby tests. Then produce:

1. A short behavior summary grounded in the source.
2. Existing test framework and naming conventions.
3. A scenario table covering happy paths, boundaries, invalid input, and
   regression cases.
4. Property-style invariants if the behavior has stable rules worth checking.
5. Integration or contract cases only where existing project tests already use
   those patterns.
6. Mutation-score or coverage gaps only if the repository already has a tool for
   measuring them.
7. The smallest test command to run after the draft is written.

After the plan, draft the minimal tests that fit the project's current test
stack. Do not introduce a new framework unless the project has no test framework
yet and you explain the tradeoff.

Claude Code derives the command name from the filename, so this file creates /test-advanced. Text after the command name is passed through $ARGUMENTS.

Usage

/test-advanced src/utils/parser.ts

/test-advanced "the retry backoff behavior in packages/client/retry.ts"

/test-advanced apps/api/users.test.ts -- focus on regression cases from the current diff

The text after the command is normal prompt input, not a CLI parser. There are no documented built-in flags such as --mutation, --pytest, or --coverage; ask for those focuses in plain language, and only rely on tools that already exist in the repository.

Suggested workflow

1. Ground the target. Ask Claude to read the target file, adjacent tests, and package configuration before proposing new cases.
2. Define success criteria. State what behavior should stay true, which failures matter, and what a useful regression test would catch.
3. Match the project. Reuse the existing test framework, assertion style, fixture pattern, and command names.
4. Separate plan from edits. Have Claude show the scenario table first, then write the smallest useful tests.
5. Run narrowly. Use the smallest test command for the touched file or package before running the full suite.

Example output shape

Ask the command to return a compact plan before writing tests:

Then let Claude draft tests only for cases that can run in the current project. For JavaScript or TypeScript this might mean Jest or Vitest if those configs already exist. For Python this usually means pytest when the repository already uses it.

Source Verification Notes

Verified on 2026-06-19:

• Claude Code's skills/custom-command documentation says custom commands have been merged into skills, and existing .claude/commands/*.md files still create slash invocations based on the filename.
• Anthropic's testing guidance recommends defining specific, measurable success criteria and designing evaluations around representative tasks, edge cases, and regression checks.
• This page applies those mechanics and testing principles to a user-authored /test-advanced recipe. It does not claim an official built-in /test-advanced or /test command.

Duplicate Check

/generate-tests is the simpler broad test-generation recipe. /test-advanced is narrower: it asks Claude to plan a source-grounded test matrix first and only then draft tests that fit the repository's existing framework and tooling.

Troubleshooting

• If the output invents a framework, tighten the prompt: "Use only the test framework already configured in this repository."
• If the plan is too broad, ask for the smallest regression-focused case first.
• If the command cannot find a test script, point it at package.json, pyproject.toml, pytest.ini, vitest.config.*, or the repository docs.

• Source repositorygithub.com 2026-07-13T15:19:32-07:00
• Documentationcode.claude.com