Add broken claude implementation, CI / CD testing & some other quality of life improvements

.github/workflows/ci.yml

@@ -0,0 +1,55 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  lint:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - uses: astral-sh/setup-uv@v4
+      - name: Install dependencies
+        run: uv sync --extra dev
+      - name: Ruff check
+        run: uv run ruff check .
+      - name: Ruff format check
+        run: uv run ruff format --check .
+
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - uses: astral-sh/setup-uv@v4
+      - name: Install dependencies
+        run: uv sync --extra dev
+      - name: Run tests
+        run: uv run pytest tests/ -v --tb=short
+      - name: Run live agent integration test
+        if: env.OPENROUTER_API_KEY != ''
+        continue-on-error: true
+        env:
+          OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
+        run: uv run pytest tests/test_agent_integration.py -v --tb=short -k "LiveOpenRouter"
+
+  test-with-z3:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - uses: astral-sh/setup-uv@v4
+      - name: Install dependencies
+        run: uv sync --extra dev --extra analysis
+      - name: Run analysis tests
+        run: uv run pytest tests/test_progent/test_analysis.py -v --tb=short

CLAUDE.md

@@ -0,0 +1,107 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Project Overview
+
+Progent is a **Programmable Privilege Control framework for LLM Agents**. It enforces fine-grained security policies on AI agent tool calls using JSON Schema validation. Policies are declarative (JSON) and framework-agnostic.
+
+## Common Commands
+
+### Setup
+```bash
+uv sync                          # Install all dependencies
+uv sync --extra analysis         # Include Z3 policy analysis
+uv sync --extra all              # Include all optional deps
+```
+
+### Testing
+```bash
+uv run pytest tests/             # Run all tests
+uv run pytest tests/test_core.py # Run a single test file
+uv run pytest tests/test_core.py::test_function_name  # Run a single test
+```
+
+### Linting
+```bash
+uv run ruff check .              # Lint
+uv run ruff format .             # Format
+```
+
+### Running the Example Agent
+```bash
+cd implementations/examples/coding_agent
+uv run run_agent.py              # Requires OPENROUTER_API_KEY in .env
+uv run run_agent.py --framework langchain --policies policies.json
+```
+
+## Architecture
+
+### Three-Layer Structure
+
+1. **`progent/`** — Core SDK library (pip-installable)
+   - `core.py` — Global policy/tool registry + `check_tool_call()` enforcement
+   - `policy.py` — Policy loading from JSON files or dicts
+   - `validation.py` — Argument validation (JSON Schema, regex, callables)
+   - `wrapper.py` — `@secure` decorator and LangChain tool wrapping
+   - `exceptions.py` — `ProgentBlockedError`, `PolicyValidationError`, `PolicyLoadError`, `PolicyConfigError`
+   - `analysis.py` — Z3-based conflict detection (optional, requires `z3-solver`)
+   - `generation.py` — LLM-based policy generation (optional, not yet tested)
+   - `adapters/` — `langchain.py`, `mcp.py`
+
+2. **`implementations/`** — Example agent implementations using the SDK
+   - `core/tool_definitions.py` — Single source of truth for tool schemas
+   - `core/secured_executor.py` — Tool wrapping with logging + policy
+   - `core/progent_enforcer.py` — Wrapper around progent library
+   - `frameworks/base_agent.py` — Abstract base; subclassed by `langchain_agent.py`, `adk_agent.py`, `raw_sdk_agent.py`
+   - `tools/` — Reusable tool implementations (file, command, communication)
+
+3. **`secagent/`** — Reference implementation from the Progent paper authors
+
+### Policy Format
+
+Policies map tool names to rule lists: `{tool_name: [(priority, effect, conditions, fallback), ...]}`. Effect `0` = allow, `1` = deny. Conditions are JSON Schema constraints on arguments. Default behavior is deny (no matching rule = blocked).
+
+### Policy Evaluation Flow
+
+```
+Tool Call → check_tool_call(name, kwargs)
+  → Iterate rules by priority
+  → Match allow rule? → Execute
+  → Match deny rule? → Block (fallback: 0=error, 1=terminate, 2=ask user)
+  → No match? → Block (default deny)
+```
+
+### Core API
+
+```python
+from progent import secure, load_policies, check_tool_call
+
+load_policies("policies.json")    # Load policy file
+check_tool_call("tool", {args})   # Manual check (raises ProgentBlockedError)
+
+@secure                           # Decorator
+def my_tool(arg: str) -> str: ...
+
+tools = secure([tool1, tool2])    # Wrap list of tools/functions
+```
+
+### Global State
+
+`progent/core.py` uses module-level globals (`_available_tools`, `_security_policy`, `_user_query`) with getter/setter functions. Tests must call `reset_security_policy()` in fixtures to avoid state leakage.
+
+## Key Conventions
+
+- Python >=3.11, target 3.12 (`.python-version`)
+- Ruff for linting/formatting (line-length 100, ignores E501)
+- pytest with `asyncio_mode = "auto"`
+- Build system: hatchling
+- Environment variables via `.env` (gitignored): `OPENROUTER_API_KEY`, `PROGENT_POLICY_MODEL`
+
+## Known Weak Spots (from TODO.md)
+
+- MCP adapter (`progent/adapters/mcp.py`) has no real-world testing
+- `progent/generation.py` is untested and not integrated
+- `implementations/core/progent_enforcer.py` duplicates some logic from `progent/core.py`
+- Z3 regex-to-Z3 conversion may fail on edge cases
+- No CI/CD pipeline

README.md

@@ -7,9 +7,15 @@ Secure coding agent with policy enforcement via the `progent` library.
 ### UV approach(RECOMMENDED)
 
 ```bash
-# 1. Activates the environment and installs all packages
+# 1. Install core dependencies
 uv sync
 
+# 1b. Install optional framework extras as needed
+uv sync --extra adk            # Google ADK (Gemini)
+uv sync --extra claude-sdk     # Claude Agent SDK
+uv sync --extra analysis       # Z3 policy analysis
+uv sync --extra all            # Everything
+
 # 2. Add OPENROUTER_API_KEY to your .env
 
 # 3. Run the agent
@@ -67,7 +73,8 @@ implementations/                  # AGENT IMPLEMENTATIONS
 │   ├── base_agent.py             # Shared agent logic
 │   ├── langchain_agent.py        # LangChain adapter
 │   ├── adk_agent.py              # Google ADK adapter
-│   └── raw_sdk_agent.py          # OpenAI SDK adapter
+│   ├── raw_sdk_agent.py          # OpenAI SDK adapter
+│   └── claude_sdk_agent.py      # Claude Agent SDK adapter
 ├── tools/                        # Tool implementations (stable)
 │   ├── file_tools.py
 │   ├── command_tools.py
@@ -115,17 +122,20 @@ python run_agent.py                              # Default (LangChain + sandbox)
 python run_agent.py --workspace ./my_project     # Custom workspace
 python run_agent.py --framework adk              # Use Google ADK
 python run_agent.py --framework raw_sdk          # Use raw OpenAI SDK
+python run_agent.py --framework claude_sdk       # Use Claude Agent SDK
 python run_agent.py --model anthropic/claude-3.5-sonnet  # Different model
 python run_agent.py --policies custom.json       # Custom policies file
+python run_agent.py --api-base https://api.openai.com/v1  # Use OpenAI directly
 ```
 
 | Flag | Description |
 |------|-------------|
 | `-w, --workspace` | Working directory for the agent |
-| `-f, --framework` | `langchain`, `adk`, or `raw_sdk` |
+| `-f, --framework` | `langchain`, `adk`, `raw_sdk`, or `claude_sdk` |
 | `-m, --model` | LLM model (OpenRouter format) |
 | `-p, --policies` | Path to policies JSON file |
 | `-c, --config` | Path to config YAML file |
+| `--api-base` | API base URL override |
 
 
 ## REPL Commands
@@ -144,7 +154,7 @@ All config files are in `implementations/examples/coding_agent/`:
 **`config.yaml`** - LLM settings, system prompt, logging:
 ```yaml
 llm:
-  model: meta-llama/llama-3.1-70b-instruct
+  model: deepseek/deepseek-v3.2
   api_base: https://openrouter.ai/api/v1
 ```
 

SECURITY.md

@@ -0,0 +1,34 @@
+# Security Policy
+
+We take security and privacy matters very seriously. We appreciate that our users place a high degree of confidence and trust in this project and we strive to meet those expectations.
+
+## Vulnerability Reporting
+
+If you believe you've discovered a potential vulnerability in this project's security, please contact us at echarris@smcm.edu.
+
+When reporting a potential vulnerability, please include as much of the following information as possible:
+
+* A description of the vulnerability
+* The impacted software or service and its version
+* Proof-of-concept code and/or detailed steps to reproduce
+
+## Secure Communications
+
+If you're a security researcher and you believe that you have found a security issue within this project, email the details of your findings to echarris@smcm.edu. Use PGP to protect the message by using the public PGP key below.
+
+## PGP Public Key
+
+```
+-----BEGIN PGP PUBLIC KEY BLOCK-----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+=hq5w
+-----END PGP PUBLIC KEY BLOCK-----
+```

TODO.md

@@ -21,7 +21,7 @@
 
 ## Missing Tests
 
-- [ ] Integration tests for full agent loop (user prompt → tool call → policy check → result)
+- [x] Integration tests for full agent loop (user prompt → tool call → policy check → result)
 - [ ] Tests for `progent/generation.py`
 - [ ] Tests for `progent/adapters/mcp.py` with real MCP server
 - [ ] Tests for `progent/adapters/langchain.py`
@@ -37,5 +37,5 @@
 
 - [ ] `implementations/core/progent_enforcer.py` duplicates some logic from `progent/core.py`
 - [ ] Logging is scattered - could be centralized
-- [ ] No CI/CD pipeline for running tests
+- [x] No CI/CD pipeline for running tests (`.github/workflows/ci.yml`)
 

implementations/core/__init__.py

@@ -1,14 +1,14 @@
 """Core infrastructure for the Progent coding agent."""
 
 from .logging_utils import AgentLogger, get_logger, init_logger
-from .tool_registry import ToolRegistry, tool, get_registry
 from .progent_enforcer import (
-    load_policies,
-    init_progent,
+    ProgentEnforcedRegistry,
     enforce_policy,
+    init_progent,
+    load_policies,
     wrap_tool_with_enforcement,
-    ProgentEnforcedRegistry,
 )
+from .tool_registry import ToolRegistry, get_registry, tool
 
 __all__ = [
     "AgentLogger",

implementations/core/logging_utils.py

@@ -57,10 +57,14 @@ def tool_call(self, tool_name: str, arguments: dict[str, Any]) -> None:
         json_str = json.dumps({"name": tool_name, "arguments": arguments}, indent=None)
         self.logger.info(f"LLM_TOOL_CALL: {json_str}")
 
-    def progent_decision(self, tool_name: str, arguments: dict[str, Any], allowed: bool, reason: str = "") -> None:
+    def progent_decision(
+        self, tool_name: str, arguments: dict[str, Any], allowed: bool, reason: str = ""
+    ) -> None:
         """Log Progent policy decision."""
         status = "ALLOWED" if allowed else "BLOCKED"
-        args_str = ", ".join(f'{k}="{v}"' if isinstance(v, str) else f"{k}={v}" for k, v in arguments.items())
+        args_str = ", ".join(
+            f'{k}="{v}"' if isinstance(v, str) else f"{k}={v}" for k, v in arguments.items()
+        )
         msg = f"PROGENT: {status} - {tool_name}({args_str})"
         if reason and not allowed:
             msg += f" | Reason: {reason}"
@@ -101,7 +105,9 @@ def get_logger() -> AgentLogger:
     return _logger
 
 
-def init_logger(log_dir: str = "./logs", level: str = "INFO", session_name: Optional[str] = None) -> AgentLogger:
+def init_logger(
+    log_dir: str = "./logs", level: str = "INFO", session_name: Optional[str] = None
+) -> AgentLogger:
     """Initialize the global logger with custom settings."""
     global _logger
     _logger = AgentLogger(log_dir=log_dir, level=level, session_name=session_name)