Vibe coded general agent

.pre-commit-config.yaml

@@ -1,20 +1,20 @@
 repos:
 -   repo: https://github.com/pre-commit/pre-commit-hooks
-    rev: v2.3.0
+    rev: v5.0.0
     hooks:
     -   id: check-yaml
     -   id: end-of-file-fixer
     -   id: trailing-whitespace
 -   repo: https://github.com/psf/black
-    rev: 22.10.0
+    rev: 25.1.0
     hooks:
     -   id: black
 -   repo: https://github.com/pycqa/flake8
-    rev: 7.2.0
+    rev: 7.3.0
     hooks:
     -   id: flake8
 -   repo: https://github.com/pycqa/isort
-    rev: 5.12.0
+    rev: 6.0.1
     hooks:
       - id: isort
         args: [--profile=black]

CLAUDE.md

@@ -0,0 +1,103 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Project Overview
+
+Askademic is a CLI tool that helps users find information in research papers from arXiv. It's built on PydanticAI and provides:
+- Summarization of latest papers in categories
+- Question answering by searching relevant papers
+- Specific paper retrieval by title/link/arXiv ID
+
+## Development Commands
+
+### Installation & Setup
+```bash
+pip install .                    # Install the package
+pip install -e .                 # Install in development mode
+```
+
+### Testing
+```bash
+pytest                          # Run all tests
+pytest tests/test_article.py    # Run specific test file
+python evals/evals.py           # Run evaluation tests
+```
+
+### Running the Application
+```bash
+askademic                       # Start the CLI application
+```
+
+### Code Quality
+```bash
+flake8                          # Lint code (max line length: 99)
+pre-commit run --all-files      # Run pre-commit hooks
+```
+
+## Architecture
+
+### Core Components
+
+1. **main.py**: Entry point with CLI interface using prompt_toolkit and rich console
+2. **orchestrator.py**: Main coordination agent that routes requests to specialized agents
+3. **allower.py**: Determines if user queries are scientific (routes to orchestrator or returns puns)
+4. **Specialized Agents**:
+   - **summary.py**: SummaryAgent for latest paper summaries
+   - **question.py**: QuestionAgent for Q&A with arXiv search
+   - **article.py**: ArticleAgent for specific paper retrieval and analysis
+5. **memory.py**: Manages conversation context and token limits
+6. **utils.py**: Model selection and arXiv API utilities
+7. **tools.py**: Utility functions for arXiv interactions
+
+### Agent Architecture
+- Built on PydanticAI with structured Pydantic model outputs
+- Uses orchestrator pattern - main orchestrator routes to specialized agents
+- Each agent has specific tools and capabilities
+- Memory management tracks conversation history with token limits
+
+### Model Support
+- **Gemini 2.0 Flash** (default, preferred for cost and context window)
+- **Claude 3.5 Haiku** (experimental, rate limited)
+- **Claude via AWS Bedrock** (experimental)
+- **Nova Pro via AWS Bedrock** (experimental)
+
+## Configuration
+
+### Environment Variables
+Required in `.env` file (copy from `.env-template`):
+- `LLM_FAMILY`: "gemini", "claude", "claude-aws-bedrock", or "nova-pro-aws-bedrock"
+- `GEMINI_API_KEY`: For Gemini models
+- `ANTHROPIC_API_KEY`: For Claude models
+- `AWS_ACCESS_KEY_ID`, `AWS_SECRET_ACCESS_KEY`, `AWS_REGION`: For AWS Bedrock
+
+### Key Settings
+- Max request tokens: 100,000 (configured in memory.py)
+- Usage limits: 20 requests per agent run
+- Default temperature: 0 for consistency
+- Retry strategy: 20 retries with early termination
+
+## Development Guidelines
+
+### File Structure
+- `src/askademic/`: Main package
+- `prompts/`: System prompts in separate files
+- `tests/`: Unit tests for each component
+- `evals/`: Evaluation scripts
+- `logs/`: Daily log files (auto-generated)
+
+### Testing Strategy
+- Unit tests for each agent and utility
+- Manual evaluation suite in `evals/` directory
+- CI/CD with GitHub Actions testing Python 3.11-3.13
+
+### Agent Development
+- Each agent follows PydanticAI patterns with structured outputs
+- Tools decorated with `@agent.tool` for function calling
+- Async/await throughout for API calls
+- Comprehensive logging to daily log files
+
+### Memory Management
+- Conversation history maintained with token counting
+- Automatic cleanup when limits exceeded
+- Context window management for different models

README.md

@@ -15,6 +15,7 @@ Askademic is an AI agent, working as a CLI tool, that helps you with finding inf
 * summarise the latest papers in a category
 * answer questions, searching first for relevant papers
 * retrieve info about a specific paper, by link or title
+* handle flexible academic requests that don't fit the standard categories
 
 You can also ask follow-up questions. And, it has an eye for things non-scientific... see below.
 

evals/evals.py

@@ -5,6 +5,7 @@
 from dotenv import load_dotenv
 from evals_allower import run_evals as run_evals_allower
 from evals_article import run_evals as run_evals_article
+from evals_general import run_evals as run_evals_general
 from evals_orchestrator import run_evals as run_evals_orchestrator
 from evals_question import run_evals as run_evals_question
 from evals_summary import run_evals as run_evals_summary
@@ -67,6 +68,9 @@ async def main():
         console.print("\n[bold magenta]Running article evals...[/bold magenta]")
         await run_evals_article(model_family)
 
+        console.print("\n[bold magenta]Running general agent evals...[/bold magenta]")
+        await run_evals_general(model_family)
+
 
 if __name__ == "__main__":
     asyncio.run(main())