| name | loopany-daily-followups |
| description | Use on daily cadence, at session start, or when the user says "what's due today" / "what am I forgetting" / "daily check-in". Reads today's due `check_at` items, classifies each (silent resolve / surface / defer), and brings the user only what needs attention. Every surfaced item must close before session ends. |
daily-followups — daily check-in on what's due
check_at is how loopany surfaces the thing you owe future-you. Every
day the scheduler (or the user, in a coding CLI) fires this skill.
Your job: read what's due, decide what to do with each, and bring the
user only what actually needs their attention.
When to run this
- Scheduled: once per day (agent platforms)
- On demand: user asks "what's due today," "what am I forgetting,"
"daily check-in"
- At session start (coding CLIs) — highest-leverage habit of the day
Never more than once per day. Noisy digests train the user to ignore
them.
Step 1 — Pull today's queue
loopany followups --due today
Returns artifacts whose check_at is today. Overdue sweeps belong to
[[weekly-sweep/SKILL.md]] — don't do them here.
If the list is empty, say so briefly and stop. A quiet day is a valid
outcome; don't manufacture something to surface.
Step 2 — Classify each item
Read the body:
loopany artifact get <id>
Three buckets:
A. Silently resolvable. The artifact itself answers the question
it was scheduled for. A learning whose scope obviously still holds →
extend check_at with a one-line reason. A task whose outcome is
already captured elsewhere → flip to done via
[[core-artifacts/SKILL.md]].
B. Needs the user. A task whose ## Outcome requires numbers only the
user has. A learning that may now be wrong. A signal that's come back
after being dismissed.
When the item is a learning, walk its causal chain before deciding —
the belief may rest on multi-hop evidence that's no longer fresh:
loopany trace <lrn-id> --direction backward
C. Defer with a reason. Not actionable today, but still relevant.
Push check_at forward and record why — silent deferrals rot.
Step 3 — Surface to the user
Bring only bucket B to the conversation. Short digest, one line
per item, each with:
- the artifact id or slug (so the user can refer back)
- what's being asked, in plain language
- a concrete yes/no/defer prompt
Example:
3 things need you today:
1. [tsk-…] 2-week recheck on payload-shrink — did the $/session number
settle? (expected ~$0.60)
2. [lrn-…] "short deals close 2.5x faster" — still true given Q2?
3. [sig-…] recurring signal on first-week churn, third time around.
Upgrade to a task this round?
If you silently resolved bucket A items, add one summary line at the
end: "Also extended 4 learnings whose scope still holds — no action
needed."
Step 4 — Dispatch responses
Don't handle outcomes inline. Route each response to the right skill:
- Task outcome → [[core-artifacts/SKILL.md]] § Task
- Learning revision → extend
check_at, or write a new learning with
supersedes (see [[reflect/SKILL.md]])
- Signal upgrade / dismiss → [[core-artifacts/SKILL.md]] § Signal
Step 5 — Close every item you surfaced
This is the gate that keeps the loop tight. Before the session
ends, every bucket B item must exit in one of three closed states:
- Resolved — state transition written (task flipped, signal
dismissed or upgraded, learning superseded or extended).
- Deferred —
check_at pushed forward with --reason. "Wait
until Q2 close" is a reason. "Not yet" isn't.
- Retired —
check_at removed, with a one-line note why the
artifact no longer needs reminding.
If the user's response was ambiguous, clarify — don't guess a
transition. If the user disengaged mid-digest, default-defer only the
items you actually asked about (not the whole list), with reason "no
response this session." That still counts as a closed state: you
explicitly chose to push, next session will know why.
Why this step exists: without it, nothing actually happens.
Every surfaced item sits at its original check_at, shows up again
tomorrow, surfaces in [[weekly-sweep/SKILL.md]] next week, and trains the
user to ignore both digests. A digest whose prompts don't produce
state changes is noise.
Anti-patterns
❌ Surfacing without closing
Three prompts + three unresolved responses = worse than no digest.
Every surfaced item ends the session in one of resolved / deferred / retired. Zombie items are how this skill decays into noise.
❌ Dumping raw CLI output
loopany followups returns JSON. The user never sees it. Your job is
judgment — "these 3 of 11 need you" — not transcription.
❌ Running --due overdue here
Overdue sweeps are weekly. Mixing them into the daily digest means
nothing feels urgent. See [[weekly-sweep/SKILL.md]].
❌ Deferring without a reason
Every check_at push needs a reason. "Pushed 1 week" with no
rationale is how items rot.
❌ Manufacturing a followup because the list was empty
Quiet days are real. Report it and stop.
Quick reference
QUERY: loopany followups --due today
READ: loopany artifact get <id> for each
CLASSIFY: A=silent resolve · B=surface · C=defer with reason
SURFACE: only B · one line each · concrete prompt
DISPATCH: core-artifacts / reflect handle the action
CLOSE: every surfaced item ends resolved / deferred / retired — no zombies
