ai-counsel: Skill for Claude Code

Deliberation Tester Skill

Purpose

This skill teaches Test-Driven Development (TDD) patterns for the AI Counsel deliberation system. Follow the red-green-refactor cycle: write failing test first, implement feature, verify passing, then refactor.

When to Use This Skill

• Adding new CLI or HTTP adapters
• Implementing deliberation engine features
• Building convergence detection logic
• Adding decision graph functionality
• Extending voting or transcript systems
• Any feature that affects multi-round deliberations

Test Organization

The project has 113+ tests organized into three categories:

tests/
├── unit/              # Fast tests with mocked dependencies
├── integration/       # Tests with real CLI tools or system integration
├── e2e/              # End-to-end tests with real API calls (slow, expensive)
├── conftest.py       # Shared pytest fixtures
└── fixtures/
    └── vcr_cassettes/ # Recorded HTTP responses for replay

TDD Workflow

1. Write Test First (RED)

Before implementing any feature, write a test that will fail:

# tests/unit/test_new_feature.py
import pytest
from my_module import NewFeature

class TestNewFeature:
    """Tests for NewFeature."""

    def test_feature_does_something(self):
        """Test that feature performs expected behavior."""
        feature = NewFeature()
        result = feature.do_something()
        assert result == "expected output"

Run the test to verify it fails:

pytest tests/unit/test_new_feature.py -v

2. Implement Feature (GREEN)

Write minimal code to make the test pass:

# my_module.py
class NewFeature:
    def do_something(self):
        return "expected output"

Run the test to verify it passes:

pytest tests/unit/test_new_feature.py -v

3. Refactor (REFACTOR)

Improve code quality while keeping tests green:

• Extract duplicated logic
• Improve naming
• Optimize performance
• Add type hints

Run all tests to ensure nothing broke:

pytest tests/unit -v