메뉴
Use when configuring, testing, or explaining Qingcheng AI Telegram binding, AI-assistant chat bridging, proactive notification candidates, LLM send/skip decisions, cooldowns, quiet hours, or Telegram delivery failures.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | telegram-notification-tool |
| description | Use when configuring, testing, or explaining Qingcheng AI Telegram binding, AI-assistant chat bridging, proactive notification candidates, LLM send/skip decisions, cooldowns, quiet hours, or Telegram delivery failures. |
Telegram is an AI assistant channel, not an interview channel. A web user generates a one-time bind code, sends /bind CODE in Telegram, and then Telegram messages reuse the same ChatService pipeline as /chat, including simple/agentic routing and AI assistant file memory.
Telegram must not create interview sessions or read interview agent_state. Telegram can create and manage AI-assistant scheduled tasks through the same database-backed scheduler as the web chat.
Proactive notifications are event-driven. Current candidates include resume parse completion/failure and stale application follow-up. Do not send daily onboarding or generic check-in messages. Delivery must respect TELEGRAM_ALLOWED_CHAT_IDS; test or stale accounts outside the allowlist are skipped.
The web entry is on /chat via TelegramBindCard in the session sidebar:
GET /api/v1/telegram/status first.POST /api/v1/telegram/bind-code./bind CODE.This skill has no application-data Python script. Use worker functions, API checks, and backend tests for validation.
Use the worker functions from the API container:
docker compose exec api python -c "from app.tasks.telegram_tasks import poll_telegram_updates_once; print(poll_telegram_updates_once())"
docker compose exec api python -c "from app.tasks.telegram_tasks import run_proactive_notifications_once; print(run_proactive_notifications_once(force=True))"
curl http://localhost:8000/api/v1/telegram/status -H "Authorization: Bearer <access_token>"
curl -X POST http://localhost:8000/api/v1/telegram/bind-code -H "Authorization: Bearer <access_token>"
Required .env values:
TELEGRAM_ENABLED=true
TELEGRAM_BOT_TOKEN=
TELEGRAM_ALLOWED_CHAT_IDS=123456789,@username
TELEGRAM_DEFAULT_USER_EMAIL=admin@example.com
TELEGRAM_BIND_CODE_TTL_MINUTES=10
Telegram task commands:
/tasks: list active scheduled tasks./cancel_task <id-prefix>: cancel a task./pause_task <id-prefix>: pause a task./resume_task <id-prefix>: resume a task.Natural language messages such as 每天早上 9 点提醒我看投递状态 also go through ChatService and can create tasks. Telegram-created task results are written to the task inbox and sent back to the same Telegram chat.
The bind-code API returns code, command, expires_at, and ttl_minutes; the database stores only code_hash. Polling returns processed updates. Proactive notification tick returns sent messages. Do not print bot tokens or raw Telegram payloads.
Explain chat as: web bind code -> Telegram /bind CODE -> sticky chat_session_id -> ChatService -> Telegram reply. Mention phone commands when useful: /new, /new <message>, /current, /sessions, /use <session-prefix>, /tasks, /cancel_task, /pause_task, /resume_task, /stop. Explain proactive notifications as candidate event -> allowlist filter -> rule gate -> LLM send/skip -> Telegram delivery -> notification_events audit record.
python -m pytest backend/tests/telegram -q
python -m compileall -q backend/app
docker compose exec api python -c "from app.tasks.telegram_tasks import run_proactive_notifications_once; print(run_proactive_notifications_once(force=True))"
cd frontend
npm test
Use when inspecting AI assistant file-based long-term memory, session JSONL archives, soft consolidation, Dream updates, or assistant context isolation.
Use when running Agent golden-case evaluation, checking intent accuracy, tool-call accuracy, RAG grounding, job matching, interview follow-up policy, or resume rubric compliance.
Use when reading saved jobs, application status counts, manual follow-up notes, or the user's application tracking state.
Use when inspecting FastAPI routes, controllers, backend service boundaries, repository-backed endpoints, API surface, or route availability.
Use when inspecting or explaining Qingcheng AI chat complexity classification, prompt-template injection, SupervisorAgent routing, selected tool allowlists, or agent planning for a user message.
Use when changing or reviewing the Next.js frontend UI, Tailwind layout, chat/interview pages, resume/job/application screens, or frontend API wrappers.