Mock Telegram Testing

Mock Telegram Testing

Use this skill to exercise Roomote's Telegram integration against the checked-in mock Telegram harness. Do not invent another fake Telegram stack and do not fall back to a real Telegram bot unless the user explicitly asks for that parity test.

Telegram continuity is inferred from chat id (+ forum topic id) → active cloud job. Private bot chats can have topics when Threaded Mode is enabled in BotFather, and forum supergroups can have topics when the bot has Manage Topics rights. /new is the explicit escape hatch after a task completes. That continuity logic is the highest-value thing to test here.

Quick Reference

Step 1: Seed the database

The Telegram webhook handler attributes messages only to linked senders: it needs a telegram_user_mappings row for the mock Telegram user pointing at a real local user. Query existing state first, then seed only what's missing:

# Check existing state
PGPASSWORD=password psql -h localhost -U postgres -d development -c "SELECT telegram_user_id, telegram_chat_id, user_id FROM telegram_user_mappings LIMIT 5;"

# Get a valid user for seeding
PGPASSWORD=password psql -h localhost -U postgres -d development -c "SELECT id FROM users LIMIT 1;"

Then seed (adjust user_id from the query above):

-- Link mock Telegram user 111000111 to a local Roomote user
INSERT INTO telegram_user_mappings (telegram_user_id, telegram_chat_id, telegram_username, user_id)
VALUES ('111000111', '111000111', 'grace_mock', '<user_id>')
ON CONFLICT ON CONSTRAINT telegram_user_mappings_unique DO NOTHING;

Messages from unlinked senders are acked and dropped (telegram_sender_not_linked) — that is itself a scenario worth testing, but every task-entry scenario requires the mapping above.

Step 2: Wire the API server env

The API resolves Telegram credentials from real env vars first (resolveTelegramRuntimeCredentials, cached ~30s). Set these in the API server's environment (or .env.local) before it starts:

R_TELEGRAM_BOT_TOKEN=7000000001:mock-telegram-token   # any value; the harness accepts all tokens unless acceptedBotTokens is set
R_TELEGRAM_WEBHOOK_SECRET=<any-shared-secret>         # harness reads the same var via dotenvx, so signatures match automatically
TELEGRAM_API_BASE_URL=http://127.0.0.1:3013           # reroutes ALL outbound Bot API calls to the harness

Roomote resolves roomote_mock_bot from the harness's getMe response using the configured token; no separate username variable is needed.

Without R_TELEGRAM_WEBHOOK_SECRET the webhook returns 503; with a mismatched secret it returns 401.

Step 3: Create a scenario file

Copy the example and fix the webhook target to point at the sandbox API (port 3001, not 4000):

cp packages/communication/scripts/mock-telegram.example.json /tmp/mock-telegram-test.json
sed -i '' 's|localhost:4000|localhost:3001|g' /tmp/mock-telegram-test.json   # macOS; drop '' on Linux
grep webhookUrl /tmp/mock-telegram-test.json
# Should show: "webhookUrl": "http://localhost:3001/api/webhooks/telegram"

For custom scenarios, edit /tmp/mock-telegram-test.json directly. Never mutate the committed example.

Step 4: Start the harness

pnpm --filter @roomote/communication mock:telegram --state /tmp/mock-telegram-test.json

The harness starts on port 3013, replays any events in the replay array, and keeps listening. For one-shot replay that exits after: add --exit-after-replay.

Step 5: Inject updates manually (optional)

update_id is minted automatically (unique per run) when omitted — pass an explicit updateId only when testing the Redis dedup path. entities are computed from the text (bot commands, @mentions) exactly like real Telegram, so /new and @roomote_mock_bot are detected without hand-counting offsets.

# Private-chat task entry
curl -s -X POST http://127.0.0.1:3013/mock/events \
  -H 'Content-Type: application/json' \
  -d '{
    "kind": "message",
    "message": {
      "chat": { "id": 111000111, "type": "private", "first_name": "Grace" },
      "from": { "id": 111000111, "first_name": "Grace", "username": "grace_mock" },
      "text": "what file handles Telegram webhooks?"
    }
  }'

# Follow-up while the job is active (same chat id = same conversation)
curl -s -X POST http://127.0.0.1:3013/mock/events \
  -H 'Content-Type: application/json' \
  -d '{
    "kind": "message",
    "message": {
      "chat": { "id": 111000111, "type": "private", "first_name": "Grace" },
      "from": { "id": 111000111, "first_name": "Grace", "username": "grace_mock" },
      "text": "also check the retry logic please"
    }
  }'

# Cancel button click (callback_query) — use a real cloudJobId from the DB
curl -s -X POST http://127.0.0.1:3013/mock/events \
  -H 'Content-Type: application/json' \
  -d '{
    "kind": "callback_query",
    "callbackQuery": {
      "id": "cbq-manual-1",
      "from": { "id": 111000111, "first_name": "Grace", "username": "grace_mock" },
      "data": "cancel_task:<cloudJobId>",
      "message": { "message_id": 42, "chat": { "id": 111000111, "type": "private" } }
    }
  }'

Step 6: Inspect results

Always check the mock state after replay — do not declare success just because the harness returned 200:

# Full state dump
curl -s http://127.0.0.1:3013/mock/state | jq .

# Just the messages (bot messages have from.is_bot == true)
curl -s http://127.0.0.1:3013/mock/state | jq '.messages'

# Bot replies that quote a user message (reply anchoring)
curl -s http://127.0.0.1:3013/mock/state | jq '.messages[] | select(.reply_to_message_id != null)'

# Ack reactions on inbound messages (eyes = picked up)
curl -s http://127.0.0.1:3013/mock/state | jq '.messages[] | select(.reactions | length > 0)'

# Callback answers (button spinners stopped)
curl -s http://127.0.0.1:3013/mock/state | jq '.callbackAnswers'

Scenario Selection

See references/scenarios.md for the full user-journey catalog. Common picks:

• private-fast-answer — an ordinary private message answered inline without a cloud job
• private-task-entry — DM task kickoff that creates a cloud job
• followup-to-active-job — second message in the same chat queues to the running job instead of launching a new task
• new-task-command — /new <req> after completion forces a fresh task instead of snapshot resume
• group-mention-gating — group messages ignored unless the bot is mentioned
• duplicate-delivery — same update_id twice → exactly-once handling
• unlinked-sender — message from an unmapped Telegram user is nudged and dropped
• long-reply-chunking — worker reply over 4096 chars splits into multiple messages
• cancel-button — callback_query cancels the running job
• routing-confirmation — low-confidence route posts a workspace picker; Never mind is the zero-side-effect live-test path

LLM-judged eval scenarios live in packages/communication/evals/scenarios/ and run via pnpm --filter @roomote/communication eval:telegram-scenario.

Guardrails

• Do not create a second mock Telegram server. Use the harness in packages/communication/.
• Do not use a real Telegram bot unless the user explicitly asks for that.
• Do not declare success because the harness started. Always inspect /mock/state.
• Do not mutate the committed example scenario — copy it to /tmp/ first.
• Do not assume the example webhook target port is correct. The sandbox API runs on 3001.
• Do not reuse explicit updateId values across runs — Redis dedups update ids for 5 minutes, silently dropping your event. Omit updateId unless the dedup path is what you are testing.
• Do not claim continuity behavior is covered unless you observed the second message being queued to the existing job (no second task-start ack) in /mock/state and cloud_jobs.

Output Standard

End each use of this skill with:

• the scenario used and the webhook target
• the inbound updates injected (if any)
• the key messages, reactions, or callback answers observed in /mock/state
• a pass or fail judgment
• the next debugging lead if the behavior failed