feat: migrate to capabilities, add build-in guardrails, rename from pydantic-ai-middleware to pydantic-ai-shields

.github/workflows/ci.yml

@@ -57,7 +57,7 @@ jobs:
         run: uv run pyright
 
       - name: MyPy
-        run: uv run mypy src/pydantic_ai_middleware tests
+        run: uv run mypy src/pydantic_ai_shields tests
 
   test:
     name: Test Python ${{ matrix.python-version }}

.github/workflows/publish.yml

@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-latest
     environment:
       name: pypi
-      url: https://pypi.org/project/pydantic-ai-middleware/
+      url: https://pypi.org/project/pydantic-ai-shields/
     permissions:
       id-token: write
 

.gitignore

@@ -1,3 +1,4 @@
+notes/
 # Byte-compiled / optimized / DLL files
 __pycache__/
 *.py[cod]

AGENTS.md

@@ -4,83 +4,50 @@ Instructions for AI coding assistants working on this repository.
 
 ## Project Overview
 
-**pydantic-ai-middleware** is a simple middleware library for [pydantic-ai](https://ai.pydantic.dev/) agents. It provides clean before/after hooks at every lifecycle stage without imposing a guardrails structure - you decide what to do (logging, guardrails, metrics, transformations).
+**pydantic-ai-shields** provides guardrail capabilities for [pydantic-ai](https://ai.pydantic.dev/) agents. Built on pydantic-ai's native capabilities API (v1.71+). No middleware wrappers — pure capabilities.
 
 ## Quick Reference
 
 | Task | Command |
 |------|---------|
-| Install | `make install` |
-| Test | `make test` |
-| Test + Coverage | `uv run coverage run -m pytest && uv run coverage report` |
-| Lint | `uv run ruff check .` |
-| Format | `uv run ruff format .` |
-| Typecheck | `uv run pyright` and `uv run mypy .` |
-| All checks | `make all` |
-| Build docs | `make docs-serve` |
+| Test | `uv run pytest tests/ -v` |
+| Test + Coverage | `uv run coverage run -m pytest tests/ && uv run coverage report --fail-under=100` |
+| Lint | `uv run ruff check src/ tests/` |
+| Typecheck | `uv run pyright src/` |
 
 ## Architecture
 
 ```
-pydantic_ai_middleware/
-├── base.py        # AgentMiddleware - base class with lifecycle hooks
-├── agent.py       # MiddlewareAgent - wraps agents with middleware
-├── toolset.py     # MiddlewareToolset - wraps toolsets for tool call interception
-├── decorators.py  # @before_run, @after_run, etc. - function decorators
-├── exceptions.py  # InputBlocked, ToolBlocked, OutputBlocked
-└── __init__.py    # Public API exports
+src/pydantic_ai_shields/
+  __init__.py       — Package exports
+  guardrails.py     — 5 capability implementations + exceptions + CostInfo
+tests/
+  test_guardrails.py — All tests
 ```
 
-### Key Design: Middleware Chain
+### Capabilities
 
-Middleware executes in order for `before_*` hooks and reverse order for `after_*` hooks:
-
-```python
-agent = MiddlewareAgent(
-    agent=base_agent,
-    middleware=[mw1, mw2, mw3],
-)
-# before_run: mw1 -> mw2 -> mw3 -> [Agent]
-# after_run:  [Agent] -> mw3 -> mw2 -> mw1
-```
-
-### Lifecycle Hooks
-
-| Hook | When Called | Can Modify |
-|------|-------------|------------|
-| `before_run` | Before agent starts | Prompt |
-| `after_run` | After agent finishes | Output |
-| `before_model_request` | Before each model call | Messages |
-| `before_tool_call` | Before tool execution | Tool arguments |
-| `after_tool_call` | After tool execution | Tool result |
-| `on_error` | When error occurs | Exception |
+| Capability | Hooks Used | Purpose |
+|------------|-----------|---------|
+| `CostTracking` | `before_run`, `after_run` | Token/USD tracking, budget enforcement |
+| `ToolGuard` | `prepare_tools`, `before_tool_execute` | Block tools, require approval |
+| `InputGuard` | `before_run` | Validate user input |
+| `OutputGuard` | `after_run` | Validate model output |
+| `AsyncGuardrail` | `wrap_run` | Concurrent guardrail + LLM |
 
 ## Code Standards
 
-- **Coverage**: 100% required - check with `uv run coverage report`
-- **Types**: Pyright and mypy - all functions need type annotations
-- **Style**: ruff handles formatting and linting
+- **Coverage**: 100% required
+- **Types**: Pyright strict on src/
+- **Style**: ruff for formatting and linting
 
 ## Testing
 
-Tests are in `tests/` directory. Use pytest-asyncio for async tests:
-
 ```python
 from pydantic_ai import Agent
 from pydantic_ai.models.test import TestModel
-from pydantic_ai_middleware import MiddlewareAgent, AgentMiddleware
+from pydantic_ai_shields import CostTracking, ToolGuard
 
-async def test_middleware():
-    model = TestModel()
-    model.custom_output_text = "test"
-    agent = Agent(model, output_type=str)
-    middleware_agent = MiddlewareAgent(agent, middleware=[...])
-    result = await middleware_agent.run("test")
+agent = Agent(TestModel(), capabilities=[CostTracking()])
+result = await agent.run("test")
 ```
-
-## When Modifying
-
-1. Run tests after changes: `make test`
-2. Check coverage stays at 100%
-3. Run `uv run pyright` for type errors
-4. Format with `uv run ruff format .`