Find the bugs hiding in your test suite
╔═══════════════════════════════════════╗
║ Your Code → Morris → Better Tests ║
╚═══════════════════════════════════════╝
Morris is a cargo subcommand that uses AWS Bedrock (Claude Sonnet 4.6) to perform intelligent mutation testing on Rust projects. Instead of exhaustively testing thousands of mutations, Morris uses AI to strategically select 5-8 high-value mutations that are most likely to reveal gaps in your test coverage.
Morris follows a fixed workflow — file discovery, test execution, and mutation application are all handled by deterministic code. The AI is used only for two targeted tasks: selecting which mutations to try, and analyzing the results.
┌─────────────┐ ┌──────────────┐ ┌─────────────┐
│ Your Code │ ───> │ Morris │ ───> │ Test Gaps │
│ + Tests │ │ (Fixed Flow) │ │ + Fixes │
└─────────────┘ └──────────────┘ └─────────────┘
│
├─ Discovers files (deterministic)
├─ Runs baseline tests (deterministic)
├─ AI selects mutations (Bedrock)
├─ Tests mutations (deterministic)
└─ AI analyzes results (Bedrock)
cargo install --path .- AWS Bedrock access with Claude Sonnet 4.6 enabled
- AWS credentials configured (via
~/.aws/credentialsor environment variables) - A Rust project with tests
cd your-rust-project
cargo morrisThat's it! Morris will analyze your code and report surviving mutations.
Morris uses a fixed, deterministic workflow. The AI (via AWS Bedrock Converse API) is called exactly twice: once to propose mutations, and once to analyze results.
┌─────────────────────────────────────────────────────────────┐
│ Morris Workflow │
├─────────────────────────────────────────────────────────────┤
│ │
│ 1. 📁 Discovery (deterministic) │
│ └─ Recursively finds .rs files under src/ │
│ │
│ 2. 📖 Read Sources (deterministic) │
│ └─ Reads all source files into memory │
│ │
│ 3. ⏱️ Baseline (deterministic) │
│ └─ Runs `cargo test` to verify and measure timing │
│ │
│ 4. 🧬 Mutation Plan (AI — Bedrock call #1) │
│ └─ Claude proposes 5-8 strategic mutations as JSON │
│ • Operators: > → <, + → -, == → != │
│ • Boundaries: 0 → 1, len() → len()-1 │
│ • Logic: && → ||, true → false │
│ │
│ 5. 🧪 Testing Loop (deterministic) │
│ For each mutation: │
│ ├─ Backup original file │
│ ├─ Apply mutation to single line │
│ ├─ Run tests (with 3x baseline timeout) │
│ └─ Restore original file │
│ │
│ 6. 📊 Results Summary (deterministic) │
│ └─ Counts killed / survived / build errors │
│ │
│ 7. 💡 Analysis (AI — Bedrock call #2) │
│ └─ Claude explains surviving mutations and │
│ suggests specific tests to catch them │
│ │
│ 8. ✨ Auto Mode (optional, deterministic) │
│ └─ Parses AI suggestions and writes improved tests │
│ │
└─────────────────────────────────────────────────────────────┘
| Flag | Description | Use Case |
|---|---|---|
| (none) | Default mode with Claude Sonnet 4.6 | Best quality analysis |
--quick |
Use Claude Haiku 4.5 | Faster, less thorough |
--auto |
Automatically apply test improvements | Hands-free mode |
-v / --verbose |
Enable debug logging | Troubleshooting |
# Standard analysis (recommended)
cargo morris
# Quick analysis for rapid feedback
cargo morris --quick
# Auto-apply test improvements
cargo morris --auto
# Quick + auto for maximum speed
cargo morris --quick --auto$ cargo morris
🧬 Morris v0.2.0 - AI-Powered Mutation Testing
📁 Discovering source files...
src/lib.rs
📖 Reading source files...
⏱️ Running baseline tests...
✅ Baseline passed in 1.2s (mutation timeout: 30.0s)
🧬 Asking AI for mutation plan...
Got 6 mutations
🧪 Testing mutations...
[1/1] src/lib.rs:42 - Change > to <... ❌ SURVIVED
[2/2] src/lib.rs:67 - Change + to -... ❌ SURVIVED
[3/3] src/lib.rs:89 - Change == to !=... ✅ KILLED
[4/4] src/lib.rs:23 - Change >= to >... ✅ KILLED
[5/5] src/lib.rs:51 - Change true to false... ❌ SURVIVED
[6/6] src/lib.rs:15 - Remove bounds check... 🔧 BUILD ERROR
📊 Results: 2 killed, 3 survived out of 5 testable mutations
💡 Analyzing surviving mutations...
[AI analysis with specific test suggestions]cargo-mutants is an excellent exhaustive mutation testing tool. Morris takes a different approach:
cargo-mutants — Exhaustive Approach
- Systematically generates all possible mutations
- Tests hundreds/thousands of mutations
- AST-based pattern matching
- Comprehensive coverage analysis
- Best for: CI/CD pipelines, audits
Morris — AI-Guided Approach
- Fixed workflow, AI used only for selection & analysis
- Selects 5-8 strategic mutations
- Contextual explanations of why mutations survive
- Auto-applies improvements
- Best for: Interactive development, learning
The biggest difference is that mutants is a lot more mature, and probably more useful in production code bases for now.
Morris requires AWS credentials with Bedrock access:
# Option 1: AWS CLI
aws configure
# Option 2: Environment variables
export AWS_ACCESS_KEY_ID=your_key
export AWS_SECRET_ACCESS_KEY=your_secret
export AWS_REGION=us-east-1# Enable debug logging
cargo morris -v
# Or via environment variable
RUST_LOG=debug cargo morrisMorris uses a fixed workflow with two targeted Bedrock Converse API calls. All file I/O, test execution, and mutation application is deterministic code — no agent loop or tool-use protocol.
┌──────────────────────────────────────────────────────────┐
│ Morris Architecture │
├──────────────────────────────────────────────────────────┤
│ │
│ ┌────────────┐ │
│ │ CLI │ cargo morris [--quick] [--auto] [-v] │
│ └─────┬──────┘ │
│ │ │
│ v │
│ ┌────────────────────────────────────────┐ │
│ │ Fixed Workflow Engine │ │
│ │ │ │
│ │ 1. Discover .rs files (fs) │ │
│ │ 2. Read source files (fs) │ │
│ │ 3. Run baseline tests (cargo test) │ │
│ │ 4. Get mutation plan ──────────────┐ │ │
│ │ 5. Test each mutation (cargo test) │ │ │
│ │ 6. Summarize results │ │ │
│ │ 7. Get analysis ───────────────────┤ │ │
│ │ 8. Auto-apply (optional, fs) │ │ │
│ └─────────────────────────────────────┘ │ │
│ │ │
│ v │
│ ┌─────────────────────┐ │
│ │ AWS Bedrock │ │
│ │ Converse API │ │
│ │ • Sonnet 4.6 │ │
│ │ • Haiku 4.5 │ │
│ │ (2 calls total) │ │
│ └─────────────────────┘ │
│ │
└──────────────────────────────────────────────────────────┘