Telegram captcha alert I receive — actual message

🔒 Tistory captcha detected
Time: 2026-05-10 14:22 KST
Post: AI Update — Claude Mythos Preview...
Session: jessyt

[Image attached — captcha]

Reply with just the captcha text.
No reply within 30 min → auto-abort + draft preserved.

My interview polling — actual message flow

🤖 BOT: Today's topic — "Running 30 Claude Code slash commands"
        Q 1/6: What pushed you to make your first slash command?

👤 ME : Got tired of copy-pasting the same prompt every time. Once I
        forgot a line and the output came back completely different.

🤖 BOT: Q 2/6: How many are you running now? Any naming convention?
        ...

Interview answers → first-person draft conversion prompt

# interview-to-draft.md (Claude Code slash command)
---
description: Convert Telegram interview answers into a first-person draft
---

Take the interview Q&A below and write:
1. Don't quote answers verbatim — restructure as first-person flow
2. Match voice.md tone ("I tried ~", "I found ~")
3. CLAUDE.md soft rules — 80%+ first-person / cite external sources
4. Apply blog-style-guide §A~§Q
5. Save HTML to drafts/ + send Telegram preview alert

My alert taxonomy — policy table

Event                    Alert policy             Why
─────────────────────────────────────────────────────────────────
Job failure (publish)    Batched 2x/day           I sleep at 3am
Captcha detected         Instant                  Reply needed in 30 min
Draft saved              Instant                  User review + publish needed
Success / progress       No alert                 "Still running" = noise
Captcha quota near       Instant + pause publish  Block captcha pile-up
Session corrupted        Instant + alert channel  Recovery procedure needed

First alert module — skeleton Claude laid in

#!/usr/bin/env python3
# lib/telegram_send.py — alert module dropped in from the first one-liner

import os, requests

def send(message: str, category: str = "blog") -> None:
    token = os.environ['TELEGRAM_BOT_TOKEN']
    chat_id_map = {
        'blog':     os.environ.get('TELEGRAM_CHAT_BLOG'),
        'alert':    os.environ.get('TELEGRAM_CHAT_ALERT'),
        'briefing': os.environ.get('TELEGRAM_CHAT_BRIEFING'),
    }
    chat_id = chat_id_map.get(category) or os.environ['TELEGRAM_CHAT_ID']
    requests.post(
        f'https://api.telegram.org/bot{token}/sendMessage',
        json={'chat_id': chat_id, 'text': message},
        timeout=10,
    )

My reply command router — /publish /skip /schedule

# lib/telegram_bot.py — publish review reply handling (excerpt)
COMMAND_ROUTER = {
    '/publish':    handle_publish_now,      # Publish immediately
    '/skip':       handle_skip_draft,       # Keep as draft, don't publish
    '/cancel':     handle_cancel_draft,     # Delete post + return to queue
    '/schedule':   handle_schedule,         # /schedule 14:30 → schedule at 14:30
}

def handle_telegram_reply(msg):
    """30-min timeout polling — review possible even while out."""
    cmd = msg.get('text', '').strip().split()[0]
    handler = COMMAND_ROUTER.get(cmd)
    if handler:
        return handler(msg)
    # No command match → treat as captcha reply
    return handle_captcha_answer(msg)

My bot feature evolution timeline

Day 1   Alert module (publish success/failure alerts)
Day 7   Captcha image relay + reply handling
Day 14  Interview polling loop (5–6 Qs → answers → draft)
Day 21  Batched failure reports 2x/day (kill instant alerts)
Day 30  Ops instruction reply handling (/publish /skip /schedule HH:MM)
Day 45  Quality review gate (pre-publish card + reply branch)
Day 60  Alerts compressed 7 streams → 2 (normalized)