LaodingBot (MVP)

Go-based personal Telegram Agent with:

• Telegram polling transport
• OpenAI-compatible LLM client
• SQLite conversation memory + simple compression
• Tool registry with built-in file, shell, and git tools
• Default-deny security policy via allowlists
• Soul markdown loading for bot personality
• Skills markdown loading for capability context
• ReAct decision loop with automatic tool execution

Now supports mutually exclusive message channels:

• telegram (long polling)
• feishu (official SDK websocket long connection)

Quick Start

1. Prepare env variables (see configs/env.sample).
   • The app auto-loads configs/env (or .env) if present.
   • You can also set CONFIG_ENV_FILE=/path/to/env.
   • Process environment variables override file values.
2. Choose exactly one channel with MESSAGE_CHANNEL=telegram|feishu.
   • If telegram: set TELEGRAM_BOT_TOKEN, keep FEISHU_* empty.
   • If feishu: set FEISHU_APP_ID and FEISHU_APP_SECRET, keep TELEGRAM_BOT_TOKEN empty.
3. Set log level with LOG_LEVEL=debug|info|warn|error.
   • To inspect full skill/tool execution content and detailed ReAct step traces, use LOG_LEVEL=debug.
4. Configure knowledge and reasoning:
   • SOUL_PATH for bot personality markdown.
   • SKILLS_DIR for skills markdown directory.
   • REACT_MAX_STEPS for maximum ReAct steps.
5. Create runtime directories:

mkdir -p data workspace

6. Run:

go mod tidy
go run ./cmd/bot

Telegram Usage

• Normal text enters unified agent pipeline:
  1. Receive message and load recent memory context
  2. Match relevant skill(s) from skills/
  3. If no skill matched, respond via direct LLM
  4. If skill matched, run ReAct and call tools (shell / file / git) only when needed
  5. Return final answer
• No /tool ... command is required for normal use.

Feishu Usage

• Bot uses Feishu official SDK long connection (ws) to subscribe im.message.receive_v1 text events.
• Received text is forwarded to the same agent pipeline and replied back to the same chat.

Knowledge Files

• Soul file default path: bot_context/soul.md
• Skills directory default path: skills/
• Skill format uses subdirectories: skills/<skill_name>/skill.md

Security Notes

• shell only allows commands listed in ALLOWED_COMMANDS.
• file only allows paths inside ALLOWED_DIRS.
• git only allows common git subcommands and runs inside WORK_DIR.
• Working directory for shell is limited by WORK_DIR.

Next Iteration

• Add skill runtime (process-level hot-plug via RPC)
• Add bootstrap pipeline (generate -> vet/test -> sandbox run -> register)
• Add approval gate for risky commands