chore: add Codex agent configuration

- AGENTS.md: Codex-equivalent of CLAUDE.md (project architecture,
  development directives, graphify integration rules)
- .codex/config.toml: register fallow MCP server for Codex sessions

Mirrors the Claude Code setup so the project is fully usable from
either coding assistant.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

.codex/config.toml

@@ -0,0 +1,2 @@
+[mcp_servers.fallow]
+command = "fallow-mcp"

AGENTS.md

@@ -0,0 +1,140 @@
+# AGENTS.md
+
+This file provides guidance to Codex (Codex.ai/code) when working with code in this repository.
+
+## Development Commands
+
+```bash
+# Development with watch mode
+npm run dev
+
+# Type checking and production build
+npm run build
+
+# Deploy to production
+npm run deploy
+```
+
+## Project Architecture
+
+TubeSage is an Obsidian plugin that extracts YouTube transcripts and generates structured notes using LLMs. Built with TypeScript and esbuild for cross-platform compatibility (desktop and mobile).
+
+### Core Architecture Patterns
+
+**Main Plugin Orchestration** (`main.ts`):
+- Extends Obsidian's Plugin class
+- Settings-driven configuration with comprehensive defaults
+- Dependency injection pattern for component initialization
+- Modal-based UI workflows for different processing types
+
+**LLM Factory Pattern** (`src/llm/llm-factory.ts`):
+- Creates appropriate LLM clients (OpenAI, Anthropic, Google, Ollama)
+- Lazy loading with settings invalidation
+- Unified interface across different AI providers
+
+**Cross-Platform HTTP** (`src/utils/fetch-shim.ts`):
+- Critical component that abstracts Obsidian's `requestUrl` as standard fetch API
+- Enables all LLM providers to work on both desktop and mobile
+- All HTTP requests MUST use `obsidianFetch` instead of standard fetch
+
+**Transcript Processing** (`src/youtube-transcript.ts`):
+- Static utility class for YouTube transcript extraction
+- Handles multiple YouTube URL formats and caption types
+- Robust error recovery with format fallbacks
+
+### Key Components
+
+- **Transcript Extraction**: `src/youtube-transcript.ts` - YouTube API interaction
+- **AI Integration**: `src/llm/` directory contains all LLM provider clients
+- **Utilities**: `src/utils/` - Cross-platform helpers, logging, validation
+- **Types**: `src/types/` - TypeScript definitions for external libraries
+
+### Settings Architecture
+
+The plugin uses a comprehensive settings system (`YouTubeTranscriptPluginSettings`) that drives all functionality:
+- LLM provider selection and API keys
+- Prompt customization (system and user prompts)
+- Processing parameters (temperature, max tokens)
+- File organization (folders, templates)
+- Debug and performance monitoring options
+
+### Processing Workflow
+
+1. **URL Validation** → **Video ID Extraction** → **Metadata Fetching**
+2. **Transcript Extraction** → **AI Summarization** → **Template Application**
+3. **Optional Timestamp Linking** → **File Creation**
+
+### Error Handling Strategy
+
+- Centralized error utilities with user-friendly messages
+- Provider-specific error handling in LLM clients
+- Graceful fallbacks for network and API failures
+- Debug logging throughout for troubleshooting
+
+## Critical Development Directives
+
+### **PRIMARY REQUIREMENTS - ALWAYS RESPECT:**
+
+1. **LLM Provider Abstraction**: ALL LLM providers MUST use LangChain for consistent provider abstraction
+   - Never bypass LangChain by calling provider APIs directly  
+   - Use `ChatOpenAI`, `ChatGoogleGenerativeAI`, `ChatOllama` classes from respective LangChain packages
+   - This ensures consistent interface, error handling, and maintainability across all providers
+   - All LLM calls must go through `src/llm/langchain-client.ts` using LangChain abstractions
+   - **NO DIRECT API CALLS TO LLM PROVIDERS - USE LANGCHAIN ONLY**
+   
+   **EXCEPTION - Anthropic**: Uses direct API calls with `obsidianFetch` instead of LangChain
+   - Anthropic's SDK has browser environment detection that conflicts with Obsidian's execution environment
+   - Direct API calls bypass this detection and ensure reliable operation in both desktop and mobile
+   - Must continue using direct HTTP calls to `https://api.anthropic.com/v1/messages` with custom headers
+
+2. **HTTP Abstraction Layer**: Use `obsidianFetch` from `src/utils/fetch-shim.ts` for cross-platform compatibility
+   - This is critical for cross-platform compatibility (desktop and mobile)
+   - Apply via `getLangChainConfiguration()` for providers that support custom fetch (OpenAI, Anthropic, Ollama)
+   - Some providers (Google) don't support custom fetch in constructor - use their default HTTP mechanism
+   - Never use standard fetch directly - always prefer the shim where possible
+
+3. **YouTube Scraping Method**: ALWAYS use the existing transcript extraction in `src/youtube-transcript.ts`
+   - Uses proven ScrapeCreators method via YouTube Internal API
+   - Handles multiple caption formats and error recovery
+   - Never replace with external libraries or alternative scraping methods
+   - Maintain the robust fallback mechanisms
+
+## Important Development Notes
+
+- **LLM Integration**: New providers should follow the factory pattern in `src/llm/llm-factory.ts`
+- **Cross-Platform**: Code must work on both Obsidian desktop and mobile
+- **Settings**: All configurable behavior should be driven by the settings system
+- **TypeScript**: Strict type checking is enabled - maintain type safety
+- **CSS Styling**: Avoid inline styles via JavaScript (`element.style.property`) - use CSS classes and `addClass()`/`removeClass()` instead for better theme compatibility
+
+## Build System
+
+- **esbuild** for bundling with CommonJS output
+- **TypeScript** compilation with strict settings
+- External dependencies include Obsidian API and various LLM SDKs
+- Production builds exclude sourcemaps and enable tree shaking
+
+## Development Guidelines Reference
+
+**CRITICAL: Before making ANY code changes, you MUST read the comprehensive guidelines file that follows using the Read tool:**
+[`docs/obsidian-plugin-guidelines.md`](docs/obsidian-plugin-guidelines.md)
+
+This complete reference contains all official Obsidian development guidelines including:
+- Core development principles and security
+- File system operations and data management  
+- Network requests and async operations
+- Workspace and view management
+- Performance optimization techniques
+- Modern JavaScript/TypeScript practices
+- CSS styling guidelines and anti-patterns
+- Plugin release and deployment procedures
+
+## graphify
+
+This project has a graphify knowledge graph at graphify-out/.
+
+Rules:
+- Before answering architecture or codebase questions, read graphify-out/GRAPH_REPORT.md for god nodes and community structure
+- If graphify-out/wiki/index.md exists, navigate it instead of reading raw files
+- For cross-module "how does X relate to Y" questions, prefer `graphify query "<question>"`, `graphify path "<A>" "<B>"`, or `graphify explain "<concept>"` over grep — these traverse the graph's EXTRACTED + INFERRED edges instead of scanning files
+- After modifying code files in this session, run `graphify . --update` to keep the graph current (AST-only, no API cost)