Tg — Development skill for Claude Code

`tg` keeps the runtime deliberately small: configuration, named sessions, authentication, locking, and Python process semantics. Telethon remains the Telegram API.

Telethon is the API; `tg` only provides the authenticated execution boundary. When a workflow is missing, write the missing logic as ordinary Python and run it through `tg`.

agent wants something in Telegram
        │
        ▼
      tg
        │
        ├── client.*        friendly Telethon methods
        └── functions.*     raw Telegram API when needed

**Three commands plus direct Python execution.**

tg login
tg doctor
tg skill
tg script.py

The Python distribution is `tg-harness`. The installed command is `tg`.

Give it to your agent

Install from PyPI:

uv tool install tg-harness

Or install the current GitHub version:

uv tool install git+https://github.com/speech115/tg.git

Then give the agent this instruction:

Use tg for Telegram. Read tg skill. Run tg doctor for initial setup or diagnosis,
not before every task. Ask me to handle login if authorization is missing; a busy
session is not an authorization failure. Use one tg program per decision boundary,
prefer client methods, and fall back to functions.* / types.* for raw requests.

Requires Python 3.12+ and a POSIX system (macOS or Linux).

Configure once

Create Telegram API credentials at https://my.telegram.org/apps, then create `~/.config/tg/config.toml`:

[telegram]
api_id = 123456
api_hash = "your-api-hash"

Keep the config readable only by your user:

chmod 600 ~/.config/tg/config.toml

Alternatively, set `TG_API_ID` and `TG_API_HASH` in the enviro