Initialize todos for tg-notipy operation:

{
  "todos": [
    {"content": "Parse user request and detect mode", "status": "in_progress", "activeForm": "Parsing user request and detecting mode"},
    {"content": "Verify configuration (bot token and chat ID)", "status": "pending", "activeForm": "Verifying configuration"},
    {"content": "Load mode reference and cross-reference guides", "status": "pending", "activeForm": "Loading mode reference and guides"},
    {"content": "Execute operation via scripts/tg_notify.py", "status": "pending", "activeForm": "Executing operation via tg_notify.py"},
    {"content": "Handle response and errors", "status": "pending", "activeForm": "Handling response and errors"},
    {"content": "Complete tg-notipy operation", "status": "pending", "activeForm": "Completing tg-notipy operation"}
  ]
}

Telegram Bot Notifications (Python)

Use the tg-notipy CLI. One Python script (scripts/tg_notify.py, stdlib only) sends messages, files, and albums to Telegram via the Bot API, with MarkdownV2 and HTML formatting, auto-retry on rate limits and transient errors, chat ID discovery, bot identity checks, JSON output, dry-run verification, and proxy or self-hosted base URL support. Flags may appear anywhere on the command line; one positional text argument is accepted (message text, or caption when a file flag is present, or extra album items in album mode). When no message source is given, the message is read from stdin. Each operation maps to a mode file in references/modes/. Operation-independent knowledge lives in references/guide-*.md files that modes load on demand. Field-verified failure modes live in references/gotchas.md.

Single Source of Truth

All fixed values (message and caption char limits, upload size limits, album item bounds, retry defaults, extension and MIME type maps) live in assets/constant-limits.md and assets/constant-filetypes.md. The script loads them at startup via scripts/tg_constants.py; the agent reads the same files. Prose references the constant files and never re-lists a value. Dump current values:

python3 scripts/tg_constants.py --dump all

Prerequisites

• Telegram bot created via @BotFather (bot token required)
• Bot added to target chat (private, group, or channel)
• python3 available in PATH (stdlib only, no pip dependencies)

Full environment setup and flag reference: references/guide-config.md.

Configuration Resolution

1. --token CLI flag, else TELEGRAM_BOT_TOKEN env var (or ./.env)
2. --chat-id CLI flag, else TELEGRAM_CHAT_ID env var (or ./.env)
3. Base URL: --base-url flag, else TELEGRAM_BASE_URL, else official API
4. Proxy: --proxy flag, else TELEGRAM_PROXY

A ./.env file fills token and chat ID only when they are not already set in the environment. Details: references/guide-config.md.

Gotchas (field-verified friction)

Read references/gotchas.md before executing operations. Top entries:

• G3: never re-run a send to re-check its exit status; re-running re-delivers the content as a duplicate
• G5: MarkdownV2 rejects the first unescaped reserved character with HTTP 400 naming it; prefer plain text or HTML for generated text
• G8: the local upload limit relaxes for self-hosted servers only with the --base-url flag, not TELEGRAM_BASE_URL alone

Mode Detection

Map user intent to a mode. Load the mode file, then load the guides in the Guides column before composing commands.

Cross-Reference Guides

Load a guide when its trigger matches. Guides are independent of any single operation:

Response Handling

On success the script prints Sent (message_id: <id>), Sent album (N messages), the chat ID (discover), or @username (id: N) (whoami) to stdout; with --json it prints {"ok":true,...} shapes. On failure it prints Failed: <reason> to stderr and exits 1. HTTP error codes and actions: references/guide-config.md.

BEGIN EXECUTION IMMEDIATELY

Do not ask the user what to do. Start with step 1:

1. Detect the mode from the request using the Mode Detection table.
2. Verify configuration: TELEGRAM_BOT_TOKEN is set (else ask the user), and TELEGRAM_CHAT_ID or --chat-id is available for send modes. Run --whoami first when the token validity is unknown.
3. Load the mode file for the detected mode.
4. Load every guide listed in the mode's Guides column.
5. Read references/gotchas.md and apply every relevant entry.
6. Run the command from the mode recipes and run the verification probe from the mode file. Handle the response per the mode file.

Update todos at each step. After completion, output a summary and clear with todos([]).