CLI Defaults and Project Context for Claude Code

You are the Claude Code AI assistant working symmetrically within the ToolsConnector project team. Your objective is to help the user build, debug, and maintain ToolsConnector without violating its core architectural principles.

The Core Mindset

• We are building a Primitive, not a Platform: This means focusing strictly on standardized, agnostic tool connection logic rather than business logic, dashboards, or proprietary ecosystems.
• Dual-use is critical: Your code must function perfectly for both traditional Python applications (like Django/Flask) and AI Agents (LangChain, autogen, function calling).
• Consistent Interface depth: We do not dumb down tools. If Gmail has a complex filtering feature, expose it—but do it in our consistent, standardized PaginatedList / @action pattern.

Code Quality Rules

1. Python Strategy: Modern Python features (3.9+). Prefer async-compatible patterns but ensure sync paths are robust. Fully utilize Type Hints everywhere.
2. Wrapper Tax philosophy:
   • DO NOT automatically wrap Python SDKs for every new connector unless specifically requested or if it is a major tier-1 SDK (like AWS, Google).
   • Prefer writing clean abstractions over raw HTTP when there's no SDK.
3. No Auth Implementation in the Library:
   • DO NOT write full OAuth callbacks or token storage logic for end users.
   • Instead, handle only the protocol exchange and expect developers to "Bring Their Own Key/Token" (BYOK).

Workflow Execution

• Read .agent/artifacts/ before executing tasks if an architecture plan is present.
• Utilize the skills in .agents/skills/ when executing specific tasks (e.g., implementing a new connector).
• Focus strictly on correctness, replay safety, and high-performance error handling.
• Be frugal with dependencies: Do not add torch, openai, or massive libraries to the core platform. Isolate AI-native capabilities strictly in toolsconnector-mcp or as extras_require.