update claude.md to reference mofa framework usage guide

Author: Nixxx19

.claude/CLAUDE.md

@@ -0,0 +1,112 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Git Conventions
+
+- **Commit file by file**: Always commit changes file by file to create many small, focused commits. This makes the git history easier to review and understand.
+- **Commit messages**: Keep commit messages short and lowercase. Use descriptive but concise messages.
+- **Never use Co-Authored-By**: Never add `Co-Authored-By` trailers to commit messages.
+
+## Build & Development Commands
+
+```bash
+# Build the binary
+cargo build --release
+
+# Install CLI from source
+cargo install --path cli/
+
+# Run directly
+cargo run --release -- <command>
+
+# Run tests
+cargo test
+
+# Run tests for a specific crate
+cargo test -p mofaclaw-core
+
+# Run a single test
+cargo test -p mofaclaw-core <test_name>
+
+# Format code
+cargo fmt
+
+# Lint
+cargo clippy
+
+# Check without building
+cargo check
+```
+
+## Toolchain
+
+Rust stable (1.85+) with `rustfmt` and `clippy` components (see `rust-toolchain.toml`).
+
+## Architecture
+
+mofaclaw is a Rust-based personal AI assistant with a workspace/workspace structure split into two crates:
+
+- **`core/`** (`mofaclaw-core`) — The library crate containing all core logic
+- **`cli/`** (`mofaclaw`) — The binary crate providing the `mofaclaw` CLI
+
+### Core Subsystems (`core/src/`)
+
+| Module | Purpose |
+|--------|---------|
+| `agent/` | Agent loop (`loop_.rs`), context builder (`context.rs`), subagent manager (`subagent.rs`) |
+| `tools/` | Built-in tools: filesystem, shell, web fetch/search, spawn (subagents), message |
+| `provider/` | Thin re-export layer over `mofa_sdk::llm` — no direct LLM code here |
+| `channels/` | Channel integrations: Telegram, Discord, DingTalk, Feishu, WhatsApp |
+| `bus/` | Async `MessageBus` (MPSC queue) decoupling channels from the agent loop |
+| `session/` | Conversation history via `mofa_sdk`'s JSONL-backed `SessionManager` |
+| `rbac/` | Role-Based Access Control for tools, skills, and filesystem paths |
+| `cron/` | Scheduled task execution via cron expressions |
+| `heartbeat/` | Periodic proactive wake-up service |
+| `config.rs` | Config loading from `~/.mofaclaw/config.json` |
+| `messages.rs` | `InboundMessage` / `OutboundMessage` types for bus communication |
+
+### Key Dependency: `mofa-sdk`
+
+LLM interaction, session storage, skills, and the agent loop core are delegated to the `mofa-sdk` crate (`mofa_sdk`). The `AgentLoop` wraps `mofa_sdk::llm::AgentLoop` and `TaskOrchestrator`. The `provider/mod.rs` re-exports `mofa_sdk::llm` types — no direct HTTP LLM calls exist in this codebase.
+
+**IMPORTANT**: Always use MoFA framework components (`mofa-sdk`) for LLM interactions, agent creation, and tool execution. See [MoFA Framework Usage Guide](.claude/mofa-framework-usage.md) for detailed standards and patterns.
+
+### Data Flow
+
+```
+Channel (Telegram/Discord/etc.)
+    ↓ InboundMessage
+  MessageBus (async MPSC)
+    ↓
+  AgentLoop
+    → ContextBuilder (builds system prompt from workspace files)
+    → mofa_sdk LLMAgent (LLM call)
+    → ToolRegistry (execute tool calls)
+    ↓ OutboundMessage
+  MessageBus
+    ↓
+Channel (send reply)
+```
+
+### Workspace Files (`workspace/`)
+
+The agent's personality and context come from markdown files in the configured workspace directory (default `~/.mofaclaw/workspace/`):
+- `SOUL.md` — Agent personality/values
+- `USER.md` — User profile
+- `HEARTBEAT.md` — Heartbeat/proactive message instructions
+- `AGENTS.md` — Subagent definitions
+- `TOOLS.md` — Tool configuration
+- `memory/MEMORY.md` — Persistent memory
+
+### Skills (`skills/`)
+
+Bundled skills live in `skills/<name>/SKILL.md` with YAML frontmatter. Workspace-local skills in `<workspace>/skills/` override bundled ones with the same name.
+
+### RBAC
+
+The RBAC system (`core/src/rbac/`) controls per-user access to tools and filesystem paths. Configured via `rbac` key in `~/.mofaclaw/config.json`. Channel-level `allow_from` lists provide coarse-grained access control before RBAC applies.
+
+## Config File
+
+`~/.mofaclaw/config.json` — loaded by `load_config()` in `core/src/config.rs`. Data directory: `~/.mofaclaw/data/`. Sessions stored as JSONL in `~/.mofaclaw/data/sessions/`.