docs: restore showcase and fix example links

docs/getting-started/ab-testing-servers.md

@@ -193,4 +193,4 @@ Description is clear and well-targeted.
 - [Comparing Configurations](comparing.md) — More comparison patterns
 - [Generate Reports](../how-to/generate-reports.md) — Get AI insights on your comparison
 
-> 📁 **Real Example:** [pydantic/test_10_ab_servers.py](https://github.com/sbroenne/pytest-skill-engineering/blob/main/tests/integration/pydantic/test_10_ab_servers.py) — Server version comparison and tool description impact testing
+> 📁 **Real Example:** [copilot/test_10_ab_servers.py](https://github.com/sbroenne/pytest-skill-engineering/blob/main/tests/integration/copilot/test_10_ab_servers.py) — Configuration A/B comparison

docs/getting-started/comparing.md

@@ -116,5 +116,5 @@ This helps you answer:
 - [Multi-Turn Sessions](sessions.md) — Test conversations with context
 
 > 📁 **Real Examples:**
-> - [pydantic/test_01_basic.py](https://github.com/sbroenne/pytest-skill-engineering/blob/main/tests/integration/pydantic/test_01_basic.py) — Single agent workflows
-> - [pydantic/test_04_matrix.py](https://github.com/sbroenne/pytest-skill-engineering/blob/main/tests/integration/pydantic/test_04_matrix.py) — Multi-dimension comparison
+> - [copilot/test_01_basic.py](https://github.com/sbroenne/pytest-skill-engineering/blob/main/tests/integration/copilot/test_01_basic.py) — Single-eval workflows
+> - [copilot/test_02_models.py](https://github.com/sbroenne/pytest-skill-engineering/blob/main/tests/integration/copilot/test_02_models.py) — Model comparison

docs/getting-started/iterations.md

@@ -142,4 +142,4 @@ pytest tests/ --aitest-iterations=3 --aitest-min-pass-rate=80
 - [Generate Reports](../how-to/generate-reports.md) — Report generation details
 
 > **Real Examples:**
-> - [pydantic/test_11_iterations.py](https://github.com/sbroenne/pytest-skill-engineering/blob/main/tests/integration/pydantic/test_11_iterations.py) — Iteration baseline tests
+> - [copilot/test_11_iterations.py](https://github.com/sbroenne/pytest-skill-engineering/blob/main/tests/integration/copilot/test_11_iterations.py) — Iteration baseline tests

docs/getting-started/skills.md

@@ -140,7 +140,7 @@ The report shows whether the skill improves performance.
 - [Multi-Turn Sessions](sessions.md) — Conversations with context
 
 > 📁 **Real Examples:**
-> - [pydantic/test_05_skills.py](https://github.com/sbroenne/pytest-skill-engineering/blob/main/tests/integration/pydantic/test_05_skills.py) — Skill loading, metadata, and before/after comparisons
+> - [copilot/test_05_skills.py](https://github.com/sbroenne/pytest-skill-engineering/blob/main/tests/integration/copilot/test_05_skills.py) — Skill loading and A/B comparisons
 
 ## Copilot Skills
 

docs/how-to/complete-example.md

@@ -7,7 +7,7 @@ description: "Complete end-to-end example demonstrating all pytest-skill-enginee
 This guide walks through the **hero test suite** — a comprehensive example demonstrating all pytest-skill-engineering capabilities in a single, cohesive banking scenario.
 
 !!! tip "Generate the Report"
-    Run `pytest tests/showcase/ -v --aitest-html=report.html` to generate the hero report.
+    Run `uv run python -m pytest tests/showcase/ -v --aitest-html=report.html` to generate the hero report.
 
 ## The Scenario: Personal Finance Assistant
 
@@ -24,25 +24,23 @@ This realistic scenario lets us test how well an LLM can understand and coordina
 tests/showcase/
 ├── test_hero.py           # The comprehensive test suite
 ├── conftest.py            # Shared fixtures
-├── agents/                # Agent instruction files for comparison
-│   ├── concise.agent.md
-│   ├── detailed.agent.md
-│   └── friendly.agent.md
+├── instructions/          # System prompt files for comparison
+│   ├── concise.md
+│   ├── detailed.md
+│   └── friendly.md
 └── skills/
     └── financial-advisor/ # Domain knowledge skill
-        ├── SKILL.md
-        └── references/
-            └── budgeting-guide.md
+        └── SKILL.md
 ```
 
 ## Running the Hero Tests
 
 ```bash
 # Run all showcase tests with HTML report
-pytest tests/showcase/ -v --aitest-html=docs/demo/hero-report.html
+uv run python -m pytest tests/showcase/ -v --aitest-html=docs/demo/hero-report.html
 
 # Run a specific test class
-pytest tests/showcase/test_hero.py::TestModelComparison -v
+uv run python -m pytest tests/showcase/test_hero.py::TestModelComparison -v
 ```
 
 ## 1. Basic Tool Usage
@@ -187,36 +185,26 @@ The report automatically generates a **model comparison table** showing:
 - Token usage and costs
 - AI-generated recommendations
 
-## 5. Agent Instruction Comparison
+## 5. System Prompt Comparison
 
 Compare how different agent instruction styles affect behavior.
 
-Store each style as an `.agent.md` file:
+Store each system prompt style as a Markdown file:
 
 ```
 tests/showcase/
-└── agents/
-    ├── concise.agent.md
-    ├── detailed.agent.md
-    └── friendly.agent.md
+└── instructions/
+    ├── concise.md
+    ├── detailed.md
+    └── friendly.md
 ```
 
-```markdown title="agents/concise.agent.md"
----
-name: AGENT_CONCISE
-description: Brief, to-the-point financial advice
----
-
+```markdown title="instructions/concise.md"
 You are a personal finance assistant. Be concise and direct.
 Give specific numbers and actionable advice in 2-3 sentences.
 ```
 
-```markdown title="agents/detailed.agent.md"
----
-name: AGENT_DETAILED
-description: Thorough financial analysis with explanations
----
-
+```markdown title="instructions/detailed.md"
 You are a personal finance assistant. Provide comprehensive analysis.
 Explain your reasoning, show calculations, and consider multiple scenarios.
 ```
@@ -226,20 +214,19 @@ Then parametrize tests over them:
 ```python
 from pathlib import Path
 from pytest_skill_engineering.copilot import CopilotEval
-from pytest_skill_engineering.core.evals import load_custom_agent
 
-AGENT_PATHS = list((Path(__file__).parent / "agents").glob("*.agent.md"))
+INSTRUCTION_PATHS = list((Path(__file__).parent / "instructions").glob("*.md"))
 
 
-class TestPromptComparison:
-    """Compare how different agent instruction styles affect financial advice."""
+class TestSystemPromptComparison:
+    """Compare how different system prompts affect financial advice."""
 
-    @pytest.mark.parametrize("agent_path", AGENT_PATHS, ids=lambda p: p.stem.replace(".agent", ""))
-    async def test_advice_style_comparison(self, copilot_eval, llm_assert, agent_path):
+    @pytest.mark.parametrize("instruction_path", INSTRUCTION_PATHS, ids=lambda p: p.stem)
+    async def test_advice_style_comparison(self, copilot_eval, instruction_path):
         """Compare concise vs detailed vs friendly advisory styles."""
         agent = CopilotEval(
-            name=agent_path.stem,
-            custom_agents=[load_custom_agent(agent_path)],
+            name=instruction_path.stem,
+            instructions=instruction_path.read_text(),
         )
 
         result = await copilot_eval(

docs/how-to/test-cli-tools.md

@@ -268,4 +268,4 @@ CLIServer(
 )
 ```
 
-> 📁 **Real Example:** [pydantic/test_09_cli.py](https://github.com/sbroenne/pytest-skill-engineering/blob/main/tests/integration/pydantic/test_09_cli.py) — CLI server testing with ls and cat commands
+> 📁 **Real Example:** [copilot/test_09_cli.py](https://github.com/sbroenne/pytest-skill-engineering/blob/main/tests/integration/copilot/test_09_cli.py) — CLI workflow testing

tests/integration/README.md

@@ -1,121 +1,77 @@
-# Integration Tests
+# Copilot Integration Tests
 
-These tests verify pytest-skill-engineering works with real LLM providers. Tests are split into two harnesses — **pydantic** (Eval + eval_run, BYOM) and **copilot** (CopilotEval + copilot_eval, GitHub Copilot SDK).
+These tests validate pytest-skill-engineering with real GitHub Copilot models and tools. They use `CopilotEval` with the `copilot_eval` fixture; no mocked LLM execution is used.
 
 ## Structure
 
 ```text
 tests/integration/
-├── conftest.py           # Shared constants and server fixtures
-├── agents/               # .agent.md test fixtures
-│   ├── banking-advisor.agent.md
-│   ├── todo-manager.agent.md
-│   └── minimal.agent.md
-├── pydantic/             # Eval + eval_run tests (Azure/OpenAI, BYOM)
-│   ├── conftest.py
-│   ├── test_01_basic.py          # Single eval, basic MCP tool calls
-│   ├── test_02_models.py         # Model comparison (parametrize)
-│   ├── test_03_prompts.py        # System prompt comparison
-│   ├── test_04_matrix.py         # Model × prompt 2×2 grid
-│   ├── test_05_skills.py         # Skill loading + skill-enhanced behavior
-│   ├── test_06_sessions.py       # Multi-turn sessions
-│   ├── test_07_clarification.py  # ClarificationDetection feature
-│   ├── test_08_scoring.py        # llm_score + ScoringDimension
-│   ├── test_09_cli.py            # CLIServer wrapping shell commands
-│   ├── test_10_ab_servers.py     # A/B server comparison
-│   ├── test_11_iterations.py     # --aitest-iterations=N reliability
-│   └── test_12_custom_agents.py  # Eval.from_agent_file + load_custom_agent
-├── copilot/              # CopilotEval + copilot_eval tests (GitHub Copilot SDK)
-│   ├── conftest.py
-│   ├── test_events.py            # SDK event capture
-│   ├── test_01_basic.py          # File create + refactor
-│   ├── test_02_models.py         # Model comparison
-│   ├── test_03_instructions.py   # Instruction differentiation + excluded_tools
-│   ├── test_05_skills.py         # Skill A/B comparison
-│   └── test_12_custom_agents.py  # Custom agents + forced subagent dispatch
-└── prompts/              # Plain .md system prompt files
-└── skills/               # Test skill directories
+├── conftest.py                  # Shared system prompts and constants
+├── agents/                      # Custom agent test fixtures
+├── prompts/                     # System prompt test fixtures
+├── skills/                      # Skill test fixtures
+└── copilot/
+    ├── conftest.py              # Copilot models, authentication, and limits
+    ├── test_events.py           # SDK event capture
+    ├── test_01_basic.py         # Basic file creation and refactoring
+    ├── test_02_models.py        # Model comparison
+    ├── test_03_instructions.py  # System prompt and tool filtering
+    ├── test_05_skills.py        # Skill A/B comparison
+    ├── test_06_sessions.py      # Multi-turn sessions
+    ├── test_07_clarification.py # Clarification detection
+    ├── test_08_scoring.py       # LLM scoring
+    ├── test_09_cli.py           # CLI workflows
+    ├── test_10_ab_servers.py    # Configuration A/B comparison
+    ├── test_11_iterations.py    # Iteration reliability
+    ├── test_12_custom_agents.py # Custom agent dispatch
+    ├── test_13_plugins.py       # Plugin discovery and loading
+    ├── test_14_skill_evals.py   # Skill eval execution
+    ├── test_15_skill_refinement.py # Skill refinement
+    ├── test_16_skill_benchmark.py  # Skill benchmarking
+    └── test_17_plugin_skill_workflow.py # End-to-end plugin skill workflow
 ```
 
 ## Quick Start
 
-### Pydantic harness (Azure OpenAI)
-
 ```bash
-# Prerequisites
-az login
-export AZURE_API_BASE=https://your-resource.cognitiveservices.azure.com
-# Run all pydantic tests
-uv run python -m pytest tests/integration/pydantic/ -v
+# Authenticate once
+gh auth login
+
+# Run all Copilot integration tests
+uv run python -m pytest tests/integration/copilot/ -v
 
 # Run a specific file
-uv run python -m pytest tests/integration/pydantic/test_01_basic.py -v
+uv run python -m pytest tests/integration/copilot/test_01_basic.py -v
 
 # Run a specific test
-uv run python -m pytest tests/integration/pydantic/test_01_basic.py::TestBankingBasic::test_balance_check_and_transfer -v
-```
-
-### Copilot harness (GitHub Copilot SDK)
-
-```bash
-# Prerequisites
-uv sync --extra copilot
-gh auth login
-# Run all copilot tests
-uv run python -m pytest tests/integration/copilot/ -v
+uv run python -m pytest \
+  tests/integration/copilot/test_01_basic.py::TestBasicFileCreation::test_create_python_file -v
 ```
 
-> **CRITICAL:** Never mix harnesses in one session. The plugin raises `pytest.UsageError` if both `eval_run` and `copilot_eval` are collected together.
-
 ## Prerequisites
 
-1. **Azure login** (Entra ID auth — no API keys needed):
-
-   ```bash
-   az login
-   export AZURE_API_BASE=https://your-resource.cognitiveservices.azure.com
-   ```
-
-1. **Models available** (checked 2026-02-23):
-
-   - `gpt-5.4-mini` — cheapest, use for most tests
-   - `gpt-5.5` — primary summary/default model
-   - `gpt-4.1` — alternative provider-backed model
-
-1. **For Copilot tests only:**
-
-   ```bash
-   uv sync --extra copilot
-   gh auth login  # or set GITHUB_TOKEN
-   ```
-
-## MCP Test Servers
-
-Built-in test servers in `src/pytest_skill_engineering/testing/`:
-
-- `banking_mcp.py`: `get_balance`, `get_all_balances`, `transfer`, `deposit`, `withdraw`, `get_transactions` for financial workflows.
-- `todo_mcp.py`: `add_task`, `complete_task`, `list_tasks`, `delete_task`, `get_task`, `update_task` for CRUD operations.
+1. GitHub Copilot authentication through `gh auth login` or `GITHUB_TOKEN`.
+2. A model available through the GitHub Copilot SDK.
+3. Dependencies installed with `uv sync --all-groups`.
 
-## Adding New Tests
+## Adding Tests
 
-Create evals inline using constants from `conftest.py`:
+Create evals inline and use the shared constants from `copilot/conftest.py`:
 
 ```python
-from pytest_skill_engineering import Eval, Provider
-from ..conftest import DEFAULT_MODEL, DEFAULT_RPM, DEFAULT_TPM, DEFAULT_MAX_TURNS
+from pytest_skill_engineering.copilot import CopilotEval
 
 
-async def test_my_feature(eval_run, banking_server):
-    agent = Eval.from_instructions(
-        "my-agent",
-        "You are a banking assistant.",
-        provider=Provider(model=f"azure/{DEFAULT_MODEL}", rpm=DEFAULT_RPM, tpm=DEFAULT_TPM),
-        mcp_servers=[banking_server],
-        max_turns=DEFAULT_MAX_TURNS,
+async def test_my_feature(copilot_eval, tmp_path):
+    agent = CopilotEval(
+        name="my-feature",
+        model="gpt-5.4-mini",
+        instructions="Create files as requested.",
+        working_directory=str(tmp_path),
     )
 
-    result = await eval_run(agent, "What's my checking balance?")
+    result = await copilot_eval(agent, "Create hello.py that prints 'hello'.")
 
     assert result.success
-    assert result.tool_was_called("get_balance")
+    assert (tmp_path / "hello.py").exists()
 ```

tests/showcase/conftest.py

@@ -0,0 +1,28 @@
+"""Shared configuration for the showcase test suite."""
+
+from __future__ import annotations
+
+import sys
+from typing import Any
+
+import pytest
+
+from tests.integration.conftest import BANKING_PROMPT
+
+
+@pytest.fixture
+def banking_mcp_servers() -> dict[str, dict[str, Any]]:
+    """Configure the built-in banking MCP server for Copilot sessions."""
+    return {
+        "banking": {
+            "command": sys.executable,
+            "args": ["-u", "-m", "pytest_skill_engineering.testing.banking_mcp"],
+            "tools": ["*"],
+        }
+    }
+
+
+@pytest.fixture
+def banking_system_prompt() -> str:
+    """Return the shared banking system prompt."""
+    return BANKING_PROMPT