caretaker
Skill49 starsChanged 3 months agoMITView on GitHub
Day-of-week rotating proactive sweep — Mon=connections / Tue=routines / Wed=skills / Thu=memories / Fri=findings / Sat=cross-cutting / Sun=weekly summary. Forces outward scan instead of inward thrash. Routine-only; load this skill into a daily routine that fires AssistantAgent. Adapted from goanna's caretaker.
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 95
Don't fix what you find unless trivial — surface it.
≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Caretaker — daily rotating focus sweep
Catches drift before it becomes urgent. Provides "what to do today" so an agent always has real outward work — never coasts on "monitoring continues."
When this fires
Daily, once per day in the morning (recommended cadence: 06:00 user-local). Different from heartbeat / reactive cycles — caretaker is one focused outward sweep per day.
Wire as a routine targeting AssistantAgent with loadSkills:
['caretaker'] and cron: '0 6 * * *' (or local equivalent).
The rotation
Each day-of-week gets a different focus. Adapted to the entities and modules in vite-flare-starter:
| Day | Focus | What to do |
| Mon | Connections | Walk mcp_connections rows. Any expired tokens? Any unused for 14+ days? Any test connections leaked into prod? Surface as findings. |
| Tue | Routines | Walk routines table via list_routines. Any routines that errored 3+ times this week? Any with no runs at all? Any whose lastRunCost exceeds budget? Surface findings. |
| Wed | Skills | Walk skills via list_skills. Any skills not loaded in 30+ days (check tool-usage stats)? Any skills with descriptions missing "Use when…"? Surface as findings. |
| Thu | Memories | memory_search for memories older than 90d with low recall_count. Candidates for archive. Surface count + IDs as a finding. |
| Fri | Findings audit | Walk findings with status='open' older than 30d. Stale opens are red flags — promote, dismiss, or extend explicitly. |
| Sat | Cross-cutting | Read findings across ALL agents this week. Look for cross-cutting patterns (same gotcha discovered by multiple agents). Promote shared patterns to learnings. |
| Sun | Weekly summary | Distil the week into one paragraph: what shipped, what landed, what's open, what's queued. Write to inbox via inbox_add with importance='medium'. |
For non-developer agents, adapt the rotation to the agent's surface area (e.g. for a customer support agent: Mon=tickets / Tue=common asks / Wed=knowledge gaps / etc).
State entity
Caretaker maintains a state entity to track rotations across runs.
Use entity_search({type: 'caretaker_state', status: 'open'}) to
find it; if missing, create one:
entity_create({
type: 'caretaker_state',
title: 'Caretaker rotation state',
status: 'open',
fields: {
lastVisitDate: '2026-05-06',
weeklySummaryLast: '2026-05-04',
rotationProgress: { ... } // per-focus pointers
}
})For the codebase / connections rotation specifically, use
fields.rotationProgress.lastConnection: 'gmail-personal' so the
next visit picks the next one in alphabetical order. Without state,
the rotation drifts back to inward thrash.
Steps
1. Determine today's focus
const dow = new Date().getDay() // 0=Sun … 6=Sat
const focus = ['weekly-summary', 'connections', 'routines', 'skills', 'memories', 'findings-audit', 'cross-cutting'][dow]2. Read your caretaker_state entity
What did you do last time on this focus? Where do you pick up?
3. Run the focus-specific sweep
Time-bounded: 15-30 min wall clock. Caretaker is surveillance, not deep-work. Find, surface, move on. Reactive cycles fix.
4. Write findings
For each thing worth surfacing:
record_finding({
body: '...',
category: 'caretaker-flag',
tags: ['caretaker', focus],
})Don't fix what you find unless trivial — surface it. Reactive cycles or curate skill decide.
5. Update caretaker_state entity
entity_update with new dates + rotation pointers.
6. Update memory + write Sunday summary if applicable
Sunday only: inbox_add a weekly summary to the user's inbox with
the week's highlights.
Anti-patterns
- Skipping when "nothing's happening" — that's exactly when caretaker fires. Quiet weeks accumulate small drift; caretaker catches it before it bloats.
- Doing deep-work during caretaker — surveillance, not deep-work. Find, surface, move on.
- No state entity — without it, the rotation drifts back to inward thrash.
- Manufacturing findings — if nothing surfaces, the right output
is "caretaker ran
<focus>; nothing to flag this week." Empty days are valid; they prove the system is working. - Caretaker eating reactive routine budget — caretaker is once-per-day, time-bounded. Don't run it inside other routines.
Output
By the end of caretaker:
- Today's focus has been swept
- Any items worth surfacing are in
findings caretaker_stateis updated- A
memory_addentry is written for the day - Sunday only: weekly summary written to inbox
- Agent has done something REAL this morning, not just coasted
Pairs with
- Reverie skill — different mechanism. Reverie is INWARD consolidation when quiet; caretaker is OUTWARD scanning on a schedule.
- Routine-health-check — caretaker reports findings; routine- health-check checks whether the caretaker routine itself is healthy.
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- tymrtn/tomorrowkit-for-minds★ 4same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository32 other files it gives its agents.
- AGENTS.mdAGENTS.md
- docs/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- agent-asks-tasksSkill
- code-reviewSkill
- compare-optionsSkill
- create-reportSkill
- csv-analyseSkill
- document-qaSkill
- draft-emailSkill
- enrich-errorSkill
- extract-structured-dataSkill
- fact-checkSkill
- git-diff-summariserSkill
- librarian-curateSkill
- meeting-notesSkill
- morning-briefSkill
- plan-taskSkill
- reflectSkill
- remember-conversationSkill
- reverieSkill
- review-code-securitySkill
- review-email-toneSkill
- review-outputSkill
- review-summary-faithfulnessSkill
- rewrite-for-audienceSkill
- route-findingSkill
- routine-health-checkSkill
- save-research-docSkill
- score-importanceSkill
- summarise-urlSkill
- web-researchSkill
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

