Tailtest

tailtest -- AI software testing for Claude Code

[](https://opensource.org/licenses/MIT) [](https://github.com/avansaber/tailtest) [](https://github.com/avansaber/tailtest/releases/latest) [](https://tailtest.com/platform/agent-edits/)

> You build. Claude builds. tailtest makes sure it works.

tailtest is the open-source AI software testing layer for Claude Code. It runs inside the build loop: every time Claude edits a file, tailtest queues the file, generates production-shaped scenarios using the R1-R15 rule layer, runs them with your existing test runner (pytest, jest, go test, etc.), and surfaces failures back to Claude within the same turn. Hook-based. Deterministic. No prompting required.

Open source (MIT), no telemetry, no SaaS account. Includes adversarial mode (R15) that biases scenario generation toward 8 breakage categories -- the mode we've used to find 16 real bugs across 47 open-source Python repositories.

Read more on tailtest.com · Platform overview · Agent-edit testing deep dive · Full docs

---

Install

claude plugin marketplace add avansaber/tailtest
claude plugin install tailtest@avansaber-tailtest

Restart Claude Code after install. No other setup required.

Update

claude plugin marketplace update avansaber-tailtest
claude plugin update tailtest@avansaber-tailtest

Uninstall

claude plugin remove tailtest@avansaber-tailtest

---

How it works

After any Claude-written file lands in your project:

1. tailtest generates scenarios that describe real business behavior -- not function signatures
2. It runs them using your existing test runner (pytest, vitest, jest, go test, and more)
3. Pass: tailtest: N scenarios -- all passed. Fail: one line surfaced, "want me to fix this?"

---

Supported languages

Python, TypeScript, JavaScript, Go, Ruby, PHP, Java, Kotlin, C#, Rust

Runners are auto-detected from pyproject.toml, package.json, go.mod, build.gradle.kts, *.csproj, *.sln, and other standard manifests.

---

Quick config

Create .tailtest/config.json in your project root to control depth:

{ "depth": "standard" }

Options: simple (2-3 scenarios), standard (5-8, default), thorough (10-15).

See tailtest.com/docs/config for all options.

---

Other tailtest variants

Same R1-R15 rule layer, same adversarial test mode, different host integration. This repo is the Claude Code variant.

• tailtest -- Claude Code plugin (hook-driven; this repo)
• tailtest-cursor -- Cursor plugin (hook-driven)
• tailtest-codex -- Codex CLI plugin (hook-driven)
• tailtest-cline -- Cline plugin (MCP-driven; reaches 8+ editors via Cline's host coverage)

See tailtest.com/demo/claude-code for a live walkthrough of this variant, or tailtest.com/comparison for a feature matrix across all four.

---

License

MIT

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: avansaber
• Source: avansaber/tailtest
• License: MIT
• Homepage: https://tailtest.com

Install and usage instructions live in the source repository linked above.