Farley Score Plugin for OpenCode

Port of the Farley Score Plugin for Claude Code, now adapted for OpenCode.

An OpenCode plugin that assesses test quality using Dave Farley's 8 Properties of Good Tests. Drop a folder or file path, and it returns a 0–10 Farley Index with per-property breakdowns.

What It Is

The Farley Score evaluates test suites against 8 properties that define test quality:

| Property | What It Checks | Why It Matters | |----------|--------------|----------------| | U Understandable | Tests read like specifications | Anyone can read and reason about them | | M Maintainable | Tests verify behavior, not implementation | Changes to internals don't break tests | | R Repeatable | Same result every time, anywhere | No flaky tests, no environmental dependencies | | A Atomic | Isolated, no shared state | Failures point to exactly one cause | | N Necessary | Every test adds unique value | No wasted execution time, no redundancy | | G Granular | Single outcome per test | Clear failure signals, easy debugging | | F Fast | Pure computation, no I/O | Fast feedback loop, TDD-friendly | | T First (TDD) | Written before implementation | Drives design, specifies intent first |

It works by analyzing test files, detecting quality signals (and anti-patterns), and calculating a weighted score based on the 8 properties.

Supported Languages

• Python (.py) — pytest, unittest
• Java (.java) — JUnit, TestNG
• JavaScript/TypeScript (.js, .ts) — Jest, Mocha, Vitest, Jasmine
• C# (.cs) — MSTest, NUnit, xUnit
• Go (.go) — Go testing

Auto-Discovery

The plugin automatically finds test files in common directories:

• tests/ (at project root)
• src/test/* (Java-style)
• Individual .py, .java, .js, .ts, .cs, .go files

How It Works

1. Analyze

Detects signals in your tests:

Positive signals (good):

• Asserts that match the test method name
• Simple, direct assertions
• No external dependencies
• No reflection/instrospection
• No Thread.sleep
• No file I/O
• No network I/O
• No time-based checks
• No tautologies

Negative signals (bad):

• Trivial tautologies (assertTrue(true))
• Mock-only tests (assert mock was called, nothing else)
• Thread.sleep / time-based tests
• File I/O operations
• Network I/O
• Reflection / instrospection
• Shared mutable state
• Implementation-coupling
• Over-specified interactions
• Framework tests (testing framework, not code)
• Cryptic test names

2. Score

• Each property gets a 0–10 score based on signal ratios
• Sigmoid scoring: diminishing returns on positive signals, dramatic penalties for negative signals
• Properties are weighted (Understandability/Maintainability are 1.5x, etc.)
• Final Farley Index is the average of all 8 property scores

3. Rate

| Farley Index | Rating | Meaning | |-------------|--------|---------| | 9.0 – 10.0 | Exemplary | Role-model test suite | | 7.5 – 8.9 | Excellent | Very good, minor gaps | | 6.0 – 7.4 | Good | Solid, with some anti-patterns | | 4.5 – 5.9 | Fair | Acceptable, but mixed quality | | 3.0 – 4.4 | Poor | Serious problems | | 0.0 – 2.9 | Critical | Fundamental issues |

Installation

One-Command Install

npx @tokmac/opencode-farley-score install

The installer runs interactively and asks:

Option 1: Project-Local (Recommended)

• Installs to .opencode/ in your current project
• Includes slash commands (/farley-score, /farley-score-coach)
• Auto-registers in opencode.json (or .opencode/opencode.json)
• Only available in this project
• Perfect for team sharing

Option 2: Global

• Installs to ~/.config/opencode/
• Available in all projects
• Plugin tools only (no slash commands)

What the Installer Does

1. Detects your project structure
2. Copies files to chosen location
3. Updates opencode.json config
4. Done — no git clone, no manual copying

After Install

Restart OpenCode or run /reload-plugins to load the plugin.

Verify

npx @tokmac/opencode-farley-score verify

Checks that files are in the right place and commands are registered.

Uninstall

npx @tokmac/opencode-farley-score uninstall

Shows exactly what will be removed, asks for confirmation, cleans up safely. Won't touch other plugins or commands.

Usage

Slash Commands

/farley-score [target]

Analyze test quality:

> /farley-score tests/
> /farley-score src/test/java/
> /farley-score test_calculator.py

/farley-score-coach [topic]

Interactive coaching:

> /farley-score-coach
> /farley-score-coach trivial-tautology
> /farley-score-coach mock-tautology

Auto-Invoked Tools

OpenCode automatically invokes the tools based on context:

> Analyze my tests in tests/
> I want to learn about test quality
> What's wrong with this test?

Example Session

Analysis:

> /farley-score tests/

🔍 Analyzing tests...
  Files: 2
  Methods: 21
  Signals: 45

📊 Farley Index: 7.2 (Good)

  U Understandable    5.0 │█████      │
  M Maintainable      4.3 │████▎      │
  R Repeatable        8.7 │████████▋  │
  A Atomic            9.5 │█████████▌ │
  N Necessary         9.1 │█████████▏ │
  G Granular          8.8 │████████▊  │
  F Fast              9.3 │█████████▎ │
  T First (TDD)       1.0 │█          │

💡 Top Issues:
  • T (First/TDD): 0/10 tests written before implementation
  • U (Understandable): 4 cryptic names
  • M (Maintainable): 3 implementation-coupled tests

  Exercise: /farley-score-coach trivial-tautology

Coaching:

> /farley-score-coach trivial-tautology

🎓 Exercise: Trivial Tautology (beginner)

Affected properties: Necessary, First (TDD)

These tests are meaningless — they pass no matter what.

Code:
  def test_true_is_true(self):
      self.assertTrue(True)

Question: What is wrong with these tests?

Hint: Would they still pass if you deleted ALL production code?

Your answer:
  It's a tautology that provides no value

✅ Feedback: Good! You identified the key issues.
  Score: 2/3
  Explanation: A test that passes even when all production code is deleted
  provides zero information. It doesn't verify behavior.
  Matched: tautology, necessary, always pass

Coaching Exercises

| Exercise | Difficulty | Properties | Anti-Pattern | |----------|-----------|------------|-------------| | trivial-tautology | beginner | N, T | Trivial Tautology | | mock-tautology | intermediate | N, M | Mock Tautology | | mock-only | intermediate | N, T | Mock-Only Test | | mega-test | beginner | G, U | Mega-Test | | shared-state | intermediate | R, A | Shared Mutable State | | sleep | beginner | F, R | Time Sleep | | implementation-coupling | advanced | M | Implementation Coupling | | over-specified | advanced | M | Over-Specified Interactions | | framework-test | beginner | N | Framework Test | | cryptic-name | beginner | U | Cryptic Test Name |

Plugin Structure

.opencode/
├── commands/
│   ├── farley-score.md          # Slash command: /farley-score
│   └── farley-score-coach.md    # Slash command: /farley-score-coach
└── plugins/
    └── farley-score/
        ├── index.js             # Plugin entry point (exports tools)
        ├── package.json         # Plugin metadata (type: module)
        └── lib/
            ├── core.js          # Math primitives (sigmoid)
            ├── scoring.js       # Farley Index calculation
            ├── analyze.js       # Test file analyzer
            └── coach.js         # Coaching exercises

Testing

Run the smoke tests:

node test/smoke.test.js

Tests cover:

• Scoring formulas (sigmoid, full pipeline, rating boundaries)
• Analyzer (file discovery, signal detection, error handling)
• Coach (exercise listing, answer validation)

Examples

The examples/ directory contains a sample project with deliberately mixed-quality tests:

• examples/sample-project/tests/test_calculator.py — 13 tests (3 good + 10 bad)
• examples/sample-project/tests/test_user_service.py — 8 tests (4 good + 4 bad)

These are used by the coaching exercises and serve as reference material.

Attribution

• Dave Farley — 8 Properties of Good Tests
• Andrea Laforgia — Farley Score methodology and signal detection
• Bernard McCarty — Plugin implementation
• Lyubomir Mitkov — Plugin port to Opencode

Port History

This is a port of the original Farley Score Plugin built for Claude Code.

Key differences:

• Pure JavaScript — Eliminated Python dependency entirely (OpenCode plugins are JS-native)
• Clack UI — Modern interactive installer with @clack/prompts
• OpenCode format — Uses tool API with proper OpenCode plugin structure
• Auto-discovery — OpenCode loads plugins from .opencode/plugins/ automatically
• Single config — Uses plugin (singular) key in opencode.json

Original: github.com/mse-online/farley_score_plugin

License

MIT