# Agent Files, Skills, and Testing

## Agent files
Agent files document how the assistant should behave, how it should interpret the repository, and how it should respond to requests. This keeps behavior consistent and auditable.

### What belongs in agent notes
- Limits and constraints (for example, streaming assumptions)
- Preferred workflows
- Any non-obvious project rules

## Skills
Skills are structured instructions that extend the agent with targeted workflows. They should be used when a request maps clearly to the skill description.

### How to use skills
- Load only what is needed
- Follow the steps, keep context small
- Prefer scripts and templates when provided

## Testing
Testing is intentionally simple and real-world.

### Local tests
- Run the local API with environment variables
- Use the smoke scripts to verify /api/token and /api/chat

### Deployed tests
- Call /api/token to obtain a JWT
- Use the streaming smoke script to verify SSE

### Expectations
- The assistant should only answer using retrieved content
- If not found, it should say: "I do not have that in my materials"

## Why this matters
Clear agent behavior and reliable test flows prevent silent regressions and make the system safer to operate.