First tasks

After installation and connecting your agent, give it a task in ordinary language. Start with reading. Logging in connects the account; it does not download every chat's history. A search through older conversations may need a separate, bounded history fetch.

Try a read-only taskCopy Anchor Link

Copy this into your agent:

Use tg cli to list five recent work chats and check my incoming messages for requests that need
an answer. Show the messages behind your conclusions and say which chats you checked. Read only:
don't send anything or mark anything read.

The agent can start with:

tg chats list --limit 5
tg inbox --limit 5

Unread messages are a starting point, not proof that someone needs an answer. The agent should inspect context when necessary and tell you about skipped chats or truncated results. Ordinary reads do not mark messages read. Reading and inbox explains their limits.

The examples below use fictional chats, IDs, messages and results. They illustrate possible dialogues, not commands you should run unchanged. Commands follow the Telegram v0.24 reference; your agent should check the installed CLI's skill and command help. MAX has its own usage reference; do not assume every Telegram option has a MAX equivalent.

Find a decision in older historyCopy Anchor Link

You

In “Atlas · team”, find the analytics fee we agreed in September 2026. Show the estimate and the
confirmation. Check the local archive first; tell me if it lacks that period. Don't send anything.

The agent checks the chat and its archive

tg chats list --search Atlas --kind group
tg store status -1001001001001
tg messages search 'analytics after:2026-09-01 before:2026-10-01' --chat -1001001001001 --json --language legacy

Agent

I found “Atlas · team”, but its local history only covers 28 September onward. The search found no fee in those messages. That doesn't tell us whether a fee was agreed earlier in the month. May I fetch this chat's history back to 1 September?

You

Yes, only that chat. Fetch back to 1 September. If more than 1,000 messages are needed, stop and
tell me what remains.

The agent fetches, checks coverage and verifies a match

tg store fetch -1001001001001 --since-time 2026-09-01 --limit 1000
tg store status -1001001001001
tg messages search 'analytics after:2026-09-01 before:2026-10-01' --chat -1001001001001 --json --language legacy
tg messages context -1001001001001 4312 --before-n 2 --after-n 4

Agent

Agreed fee: €1,500 excluding VAT. Marco proposed it on 5 September (message 4312); you confirmed it on 6 September (message 4315). The surrounding discussion describes the analytics setup, not ongoing support. The archive still has gaps, so I cannot rule out a later change elsewhere in September.

Sources: “Atlas · team”, messages 4312 and 4315. Nothing was sent.

The fee is supported by the estimate and confirmation, but a partial archive cannot establish that it remained unchanged for the rest of the month. Another bounded fetch can resume later. A fetch from September also reads newer history on its way back, while the search filters narrow the result to September.

Why these commands: messages list reads a window of messages and stores them locally; store fetch fills a larger stretch of history. messages search searches only what is stored. messages context checks what was said around a match. Neither the latest 20 messages nor an empty local search proves that a month has been checked. Archive and search.

Prepare for a meetingCopy Anchor Link

You

Prepare a 20-minute agenda for the Atlas call with Anna and Tom. Check the project and design
groups plus my direct chats with those participants over the past week. Separate decisions,
current risks and open questions. Cite sources; don't send anything.

The agent finds groups and participants, then reads the selected chats

tg chats list --search Atlas --kind group
tg contacts list --search Anna
tg contacts list --search Tom
tg messages list -1001001001001 --after-time 7d --limit 100
tg messages list -1001001001002 --after-time 7d --limit 100
tg messages list 100200300 --after-time 7d --limit 100
tg messages list 100200301 --after-time 7d --limit 100

Agent

Decisions: the launch is set for 9 October; the smaller analytics scope is approved.

Current risk: the revised design has arrived, but nobody has confirmed acceptance yet. Tom's direct message supersedes the older “waiting for the design” blocker in the group.

Agenda: confirm scope (5 minutes), approve the revised design (10 minutes), assign owners for launch checks (5 minutes).

Sources: team messages 4501/4508, design message 732, Tom's direct message 188. Checked the two groups and both direct chats for the past week; no sending or read marks.

You

Check whether Anna approved the revised design after Tom sent it. If not, add that as a question.

The agent reads the follow-up window

tg messages context -1001001001002 732 --before-n 2 --after-n 10
tg messages list 100200300 --after-time 2026-10-02T12:00:00+02:00 --limit 50

Agent

Anna acknowledged receipt but did not explicitly approve it in the messages checked. Add: “Anna, can we approve the revised design, or are changes still needed?” Keep approval open.

Sources: design message 735 and Anna's direct message 204. No reply was sent.

The sample time is the fictional design's delivery time. If a read window is truncated, the agent must follow the message cursor or report the missing part before claiming to have checked the whole week. It should ask about ambiguous participant names rather than choose a direct chat without evidence. Message windows and cursors.

Check promises and prepare repliesCopy Anchor Link

You

Check Atlas discussions from the past week for promises and questions involving me. Include
the team group and my direct chats with Anna and Tom. Separate what I owe from what I'm waiting
for, check for later completion, and draft replies. Don't send them.

The agent resolves the chats and reviews both sides of each conversation

tg chats list --search Atlas --kind group
tg contacts list --search Anna
tg contacts list --search Tom
tg review --chat -1001001001001 --since-time 7d
tg review --chat 100200300 --since-time 7d
tg review --chat 100200301 --since-time 7d

Agent

You owe Anna: comments on the slides, promised for 2 October. I found no completion in the checked messages. Sources: team message 4520, Anna's direct message 201.

Waiting for a decision: Anna's approval of the revised design. Source: design discussion quoted in Tom's direct message 188.

Completed: Tom sent the design in his direct chat. His earlier promise is no longer open.

Draft to Anna: “I've received the slides. I'll confirm a delivery time for my comments shortly.” Nothing was sent; this covers only the three chats listed above.

You

Don't invent a delivery time. Check what I actually promised Anna and make the draft specific.

The agent checks the original promise and its replies

tg messages context -1001001001001 4520 --before-n 3 --after-n 6
tg messages context 100200300 201 --before-n 2 --after-n 5

Agent

You promised comments on the pricing slides by Friday, 2 October. Draft: “I promised comments on the pricing slides by Friday and missed that deadline. I'll send you an updated delivery time.” This acknowledges the promise without making a new deadline up.

Sources: team message 4520 and Anna's direct message 201. Still a draft.

review supplies messages for the agent to interpret. Its unanswered-question filter is a heuristic, not a complete account of everyone's obligations. The agent should report incomplete review windows and check later messages before labelling an old promise outstanding. Review and unanswered questions.

Other useful requestsCopy Anchor Link

For prompts you can adapt, continue to How to phrase requests. For command syntax, use Telegram commands or MAX commands.