nodecode-checkin
Idle check-ins: after your chat has been quiet for a while, one prompt goes into the session you bound and its answer reaches you wherever that chat goes — with the wait doubling each time you leave one unanswered, and nothing firing outside your waking hours. A Nodecode rewrite of capthvnsen/alans-way-agents by capthvnsen.
Get it
Tell the agent "add the checkin cell", or run nodecode add nc://checkin.
Set it up
- Open the chat you want to be checked in on — Telegram, Discord, a browser tab, the TUI — and
send
/checkin on. That binds the session you typed it in;/checkin on SESSIONnames one instead. - Tell it your offset:
/checkin utc -300is New York in winter,+60Berlin in winter,0UTC. The rule keeps your local hours from it. /checkin statussays the wait, the hours, the offset and when the next one is due.
The same fields live in /settings under checkin (session, base_minutes, active_start,
active_end, utc_offset_minutes, paused, morning_brief, voice, tick_seconds); the
command writes them there, so a restart keeps them. A secret is not among them: the cell holds
none.
Use it
| Command | What it does |
|---|---|
/checkin or /checkin status |
the rule in one line: the wait, the hours, the offset, the pause, the next check-in |
/checkin on [SESSION] |
check in on this chat (or that session) |
/checkin off |
stop checking in on anything |
/checkin now |
one check-in now, whatever the clock says |
/checkin less / normal / more |
a check-in after 4 / 2 / 1 hours of quiet |
/checkin pause / resume |
stop and start again, keeping the rule |
/checkin hours 9-23 |
your waking hours, in your own offset |
/checkin utc -300 |
your offset from UTC, east positive |
/checkin brief on / off |
one short brief a day, just after your hours open |
/checkin voice texting / default |
how your chats read |
The model has one tool, checkin. It reads the rule with action: status, and it can only make
check-ins quieter: level less, a higher base_minutes, paused: on, an earlier
active_end or later active_start, morning_brief: off, or a new utc_offset_minutes. A
change that would reach further is refused with ask_user_to_send, and the model tells you the
exact command — only you check in more often, resume, widen the hours or turn the brief on.
What a check-in is made of is the original plugin's: one of three moves — do one safe, reversible
thing and say what it did, offer one specific thing it could do, or ask about a decision you own —
built on one open thread from its memory of your goals, one to three short sentences, and exactly
[SILENT] when nothing is worth your attention.
At each turn's first round, the cell also hands the model the texting-voice note (plain words,
no preambles, headlines rather than logs, at most three sentences for a casual reply). Set
voice to default to leave your replies in the model's own voice.
It keeps one file, checkins/state.sexp under your Nodecode home, holding timestamps and a
count: the last exchange, the last check-in, how many went unanswered, the last brief. It holds
no conversation, and it makes no network call of its own — a check-in is a prompt in a session
of yours, and its answer is a normal turn.
What is in it
| File | What it is |
|---|---|
package.lisp |
the package, the prompts as data, and the clock arithmetic |
state.lisp |
the one state file, and the section's values as the cell runs on them |
cell.lisp |
the rule, the checkin tool, /checkin, the two hooks, (help :checkin) |
test/ |
the lifecycle, the tool, the command and the rule, over a stubbed ingress and a fixed clock |
What it does not do
The original plugin's workspace skills (browser errands, a companion desktop app over SSH), its
cross-bot digest directory and cross_bot_ask tool, and its send-time polish of a leading
"Certainly!" are not carried over; a check-in fires for the one session you bind, not for several
bots. See NOTICE.
License
MIT, the original's: see LICENSE and NOTICE.
