Wallaby for coding agents | AI tests that catch real bugs

Wallaby for coding agents
AI can write low-value tests that pass without checking much. Wallaby helps your agent strengthen assertions and verify that tests catch regressions.
Run in your project and choose your coding agent.
Ask your agent to improve an existing test suite.
Your agent tests expected behavior and edge cases, then verifies its assertions.
Your agent strengthens assertions, covers missing cases, and removes redundancy.
Your agent strengthens weak assertions, adds missing cases, and removes duplicate checks.
Your agent strengthens useful tests and removes weak or duplicate checks.
Up to 40% better results with Wallaby. Measured in our internal Wallaby Bench evals.*
Your agent finds untested code.
Your agent sees which expressions each test reaches. When needed, it targets mutation testing at plausible bugs to check the strength of its assertions.
Your agent follows the test's path.
Your agent follows execution across files to check the intended behavior. When needed, traces help explain why a test caught or missed a mutation.
Your agent inspects values.
Your agent inspects actual values when a result is unclear. If a meaningful mutation survives, it can strengthen the assertion and repeat the mutation check.
Your agent checks each change.
Wallaby automatically reruns affected tests. Your agent can introduce one small source mutation, check whether the tests kill it, then restore the source and confirm they pass again.
More about Wallaby's test feedback
Writing and running tests can still leave weak assertions or tests that mirror the implementation. Wallaby helps your agent focus on high-value tests: checks of intended behavior that would fail on plausible bugs.
In our internal Wallaby Bench evals, agents using Wallaby and its skills scored up to 40% better than agents running tests through Vitest or Jest alone. This combined score covers code and test quality, speed, and token efficiency.
The evals use repeated runs, hidden checks, mutation testing, and human or AI grading against predefined criteria.
Yes. Checking that a shipping fee is defined can pass even when the amount is wrong. Wallaby helps your agent check the expected amount and test boundary cases.
Your agent checks what is already covered, adds missing cases, and can remove low-value tests that duplicate other checks while preserving distinct behavior.
Yes. Your agent can use /wallaby write and /wallaby improve in your workflows, or use the wallaby-cli skill directly to run tests and inspect coverage, traces, and runtime values.
Agents that support skills, including Claude Code, Codex, GitHub Copilot, Cursor, OpenCode, and Pi. Your agent can start Wallaby in the background or connect to an existing editor instance.
Yes. During beta, containers and cloud agents require a CLI token, including for existing license holders. A licensed Wallaby extension includes local CLI access on the same machine without a token.
The agent needs write access to ~/.wallaby, the required network access, and permission to connect to the local Wallaby process.
Wallaby supports JavaScript and TypeScript with Jest, Vitest, node:test, Mocha, Jasmine, AVA, QUnit, and Karma with Jasmine or Mocha.
Wallaby for Python is in beta, with support for pytest and unittest.
Ask your agent to improve the tests for your latest changes.
Free local CLI access during beta.
Run this in your terminal, enter your email, and follow the activation link.
A CLI token is required, even with an existing license. Generate it with the same access command.
Your existing license includes Wallaby for coding agents on this machine, with no extra activation or token needed.