killdiff
Provides a reusable GitHub Action and SARIF 2.1.0 output for GitHub Code Scanning, enabling mutation-testing gates and pull request reporting in GitHub workflows.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
killdiff
Zero-config polyglot diff-only mutation tester that catches hollow AI-generated tests in seconds. CLI, stdio MCP server (verify_diff_tests), and GitHub Action with zero runtime dependencies.
Catch Hollow AI Tests in Seconds (Zero Setup)
npx killdiff --demoSpins up a temporary git repo with a simulated refund-approval PR (src/auth.mjs + weak unit tests that pass baseline), mutates only the changed diff lines in parallel isolated sandboxes, and flags every surviving logic bug where tests failed to fail (25% score: 3 killed, 9 survived out of 12 mutants).
Related MCP server: testgen-agent
Quick Start
# Mutate only lines changed against main
npx killdiff main...HEAD
# Enforce an 80% mutation score gate in CI with Markdown PR table
npx killdiff origin/main...HEAD --max 25 --fail-under 80 --markdown
# Output SARIF 2.1.0 for GitHub Code Scanning
npx killdiff main...HEAD --sarif > killdiff.sarif
# Override test runner explicitly
npx killdiff main...HEAD --cmd "pytest -q"Connect to Your Coding Agent (Stdio MCP)
killdiff mcp exposes verify_diff_tests so Antigravity, Claude Code, Codex, Cursor, and OpenCode can verify their own newly written tests actually kill logic mutations before handing a PR back to you.
{
"mcpServers": {
"killdiff": {
"command": "npx",
"args": ["-y", "killdiff", "mcp"]
}
}
}Reusable GitHub Action
- uses: flipcyde101/killdiff@v0.1.0
with:
range: origin/main...HEAD
max-mutants: 25
fail-under: 80How killdiff Stays Fast & Safe
Diff-only scope: Parses
git diff --unified=0so only executable code on changed lines is mutated (skips test files, comments, strings, imports, and type declarations).Polyglot on Day 1: Mutates TypeScript/JavaScript, Python, Go, and Rust across boundary (
>=/>,<=/<), equality (===/!==,==/!=), logical (&&/||,and/or), boolean (true/false,True/False), and arithmetic (+/-) operators.Isolated parallel sandboxes: Never edits your real working tree. Creates ephemeral temp workspaces symlinking/junctioning
node_modules/.venv/targetfor fast setup and cleans up automatically.Baseline & timeout guards: Verifies baseline tests pass before mutating, caps runs with
--max 25, and terminates infinite-loop mutants cleanly asTIMEOUT(counted as killed).
Honest Comparison
Feature |
| StrykerJS / |
| Static AI-slop linters |
Languages | TS/JS, Python, Go, Rust | Single-language silos | Rust only | Regex/AST heuristics |
Config required | Zero config (auto-detects runner) | Per-project config file | Minimal | Minimal |
Closed-loop agent MCP ( | Yes | No | No | No |
Mathematically proves tests fail on logic bugs | Yes | Yes | Yes | No |
Where competitors win | (Lexical diff-scoped speed) | Full compiler-AST mutators across entire repositories | Deep Rust compiler integration & AST mutants | Instant millisecond feedback without running test suites |
Architecture
flowchart LR
Diff["git diff --unified=0\n(Changed lines only)"] --> Lexer["Comment & String-Aware\nPolyglot Mutator\n(TS/JS · Py · Go · Rust)"]
Lexer --> Pool["Parallel Isolated Sandboxes\n(junction/symlink deps +\n2.5x baseline timeout)"]
Pool --> Report["Terminal · Markdown PR · SARIF 2.1.0\n+ Stdio MCP (verify_diff_tests)"]Local Agent Reliability Suite
kvdiff: Zero-code local proxy pinpointing the exact UTF-8 byte breaking your LLM prompt cache (
Anthropic,OpenAI,Gemini/Antigravity,llama.cpp/vLLM).boxcheck: Local DOM geometry, CSS cascade, and layout-collision oracle (CLI + MCP) over raw Node 24 WebSocket CDP.
killdiff (this repo): Zero-config polyglot diff-only mutation tester catching hollow AI-generated tests (CLI + MCP + GitHub Action).
License
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Verification for AI coding agents: runs the tests and checks a change affects, with a receipt.
Writes adversarial test suites for AI-built code. Your agent's test engineer.
MCP server (stdio): lint OpenAPI specs with Spectral via the AgentForge API
Security reviews for coding agents: diffs checked against your org policy and live infrastructure.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceRuns a multi-agent code review pipeline via MCP, analyzing git diffs for security, quality, and logic issues and returning a pass/fail verdict.-
- AlicenseNot gradedqualityBmaintenanceProvides a code-quality backend MCP tool that automatically writes and repairs pytest tests, runs them in a sandbox, validates their effectiveness via mutation testing, and surfaces suspected source bugs for human review.MIT
- AlicenseNot gradedqualityBmaintenanceEnables agents to prove their tests catch bugs via mutation testing and to identify public capabilities no test names, so they can verify work before claiming completion.MIT
- AlicenseNot gradedqualityBmaintenanceA six-tool stdio MCP server that lets Claude Code, Codex and other agents submit a candidate change for before/after test execution and bounded StrykerJS mutation analysis inside resource-limited offline Linux containers. It returns explicit pass, fail or inconclusive verdicts bound to exact source, dependency, policy and image hashes, plus findings for removed, skipped or excluded tests.1MIT