mutation-test

Invoke the Skill with an optional path scope, for example targeting a directory like src/services. By default it tests only changed production files that have test coverage, using the native mutation runner if one is configured or the LLM-guided engine otherwise.

It detects and uses StrykerJS for TypeScript/JavaScript, Infection for PHP, mutmut for Python, PIT for JVM languages, and cargo-mutants for Rust. When no native runner is configured, it can install one with explicit consent or fall back to its LLM-guided mutation engine.

Native runners apply syntactic mutations exhaustively for a reproducible, comparable score. The LLM engine generates mutation classes native tools cannot express, such as error-path removal, state mutation, async hazards, and security guard removal. Hybrid mode reports both scores separately.

No. The Skill establishes a baseline by running the full test suite first and stops immediately if any test fails, since mutation scores against a failing suite are meaningless. It suggests repairing failing tests before proceeding.

No. Mutations are applied temporarily, executed against the relevant tests, and restored. The only lasting changes are added test assertions that close real gaps exposed by surviving mutants, plus any consented dev-dependency installs which are logged with uninstall commands.

Pass a path to scope the run, use --max N to cap total LLM mutations, --category to restrict mutation types, or --quick for at most 3 mutations per file and 20 total. The default scope is changed files only, not the whole project.