docs: add CLAUDE.md project guidance

[OriNachum]

Summary

• Adds CLAUDE.md to the repo root so Claude Code sessions have project context (architecture, commands, test markers, CI info)
• This file was previously local-only and not tracked in git

Test plan

• File committed and pushed

• Verify CLAUDE.md renders correctly on GitHub

• Confirm CI passes (no code changes)

• Claude

[qodo-code-review]

Review Summary by Qodo

Add CLAUDE.md project guidance for Claude Code sessions

📝 Documentation

Walkthroughs

Description

• Adds CLAUDE.md with comprehensive project guidance
• Documents architecture, request flow, and service layers
• Includes test commands, markers, and CI workflow info
• Provides configuration and new model integration steps

Diagram

flowchart LR
  A["CLAUDE.md Added"] --> B["Project Overview"]
  A --> C["Architecture Guide"]
  A --> D["Commands Reference"]
  A --> E["Test Markers"]
  A --> F["CI/Config Info"]
  C --> G["Request Flow"]
  C --> H["Service Layers"]
  C --> I["Model Integration"]

Loading

File Changes

1. CLAUDE.md 📝 Documentation +84/-0

Comprehensive Claude Code project guidance document

• New documentation file providing Claude Code session context
• Includes project overview, architecture, and request flow details
• Documents all test commands, pytest markers, and CI workflow
• Provides configuration variables and new model integration steps

CLAUDE.md

---

[qodo-code-review]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0) 🎨 UX Issues (0)

1. Wrong config mechanism ☑ 🐞 Bug ≡ Correctness

Description

CLAUDE.md claims config_loader uses pydantic-settings, but the implementation uses python-dotenv +
os.getenv (custom AppConfig), so readers will follow the wrong pattern when updating
configuration.

Code

CLAUDE.md[59]

+- **`utils/`** — `RequestFormatDetector` (auto-detects input format), `config_loader` (env-based config via pydantic-settings), `knowledge_base_detector`.

Evidence

CLAUDE.md explicitly states config_loader is “env-based config via pydantic-settings”, but the
actual config loader imports dotenv.load_dotenv and reads settings via os.getenv in a
hand-rolled AppConfig class; there is no pydantic-settings usage in the implementation.

CLAUDE.md[59-59]
CLAUDE.md[84-84]
src/open_bedrock_server/utils/config_loader.py[1-25]
src/open_bedrock_server/utils/config_loader.py[27-100]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

### Issue description
`CLAUDE.md` states configuration is handled via **pydantic-settings**, but the repo’s actual configuration implementation is a custom `AppConfig` that loads `.env` via `python-dotenv` and reads values using `os.getenv`. This mismatch will mislead contributors.

### Issue Context
- Current implementation: `src/open_bedrock_server/utils/config_loader.py` uses `dotenv.load_dotenv(...)` + `os.getenv(...)`.
- `CLAUDE.md` mentions pydantic-settings in multiple places.

### Fix Focus Areas
- CLAUDE.md[59-59]
- CLAUDE.md[84-84]

### Suggested fix
- Replace “via pydantic-settings” with wording that matches reality (e.g., “env-based config via python-dotenv + `AppConfig` (`os.getenv`)”).
- If pydantic-settings is intended long-term, either implement it in `config_loader` or clarify it is a dependency but not currently used.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

2. Nonexistent env var ☑ 🐞 Bug ≡ Correctness

Description

CLAUDE.md documents DEFAULT_OPENAI_MODEL as a key environment variable, but the code does not read
it and OpenAI model selection comes from request.model, so setting it has no effect.

Code

CLAUDE.md[84]

+Uses `.env` file with pydantic-settings. Key vars: `API_KEY` (server auth), `OPENAI_API_KEY`, AWS credentials (`AWS_PROFILE`/`AWS_ACCESS_KEY_ID`+`AWS_SECRET_ACCESS_KEY`/`AWS_ROLE_ARN`), `AWS_REGION`, `S3_FILES_BUCKET` (for file query feature), `DEFAULT_OPENAI_MODEL`.

Evidence

The config loader enumerates supported env vars and does not include DEFAULT_OPENAI_MODEL.
Separately, OpenAI requests pass model_id=request.model, indicating the model is taken from the
incoming request rather than any default-model env var.

CLAUDE.md[84-84]
src/open_bedrock_server/utils/config_loader.py[56-114]
src/open_bedrock_server/services/openai_service.py[244-265]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

### Issue description
`CLAUDE.md` lists `DEFAULT_OPENAI_MODEL` as a key configuration variable, but the server code does not read it. This creates a false knob for operators/contributors.

### Issue Context
- `src/open_bedrock_server/utils/config_loader.py` does not load `DEFAULT_OPENAI_MODEL`.
- `OpenAIService.chat_completion_with_request()` forwards `model_id=request.model`.

### Fix Focus Areas
- CLAUDE.md[84-84]

### Suggested fix options
1) **Documentation-only (recommended):** Remove `DEFAULT_OPENAI_MODEL` from CLAUDE.md, and clarify that the model is selected per-request via the `model` field.
2) **Feature implementation:** Add `DEFAULT_OPENAI_MODEL` support in config and request handling (e.g., when request.model is missing/empty, fall back to configured default), and ensure tests/docs reflect the behavior.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

ⓘ The new review experience is currently in Beta. Learn more

[Copilot]

Pull request overview

Adds a repo-root CLAUDE.md intended to give Claude Code sessions consistent project context (architecture, common commands, test markers, CI/config notes) for the Open Bedrock Server codebase.

Changes:

• Introduces CLAUDE.md with quick-start commands for setup, testing, linting, and running the CLI/server.
• Documents high-level architecture (routing, adapters/strategies, services) and guidance for adding new Bedrock model families.
• Summarizes pytest markers and CI workflows/configuration expectations.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.