agentleFS
derrekyoungderrekyoung/ai-chief-of-staff/.claude/skills/daily-wrap/SKILL.md

daily-wrap

Skill2 starsChanged 48 days agoMITView on GitHub

Close out the day. Ask nine reflection questions, one at a time or all at once on request, keep the answers verbatim, run a check-in on the tracker, and write it all into today's daily note. Use when the user says they are wrapping up, done for the day, ready to reflect, want to close out the day, or runs /daily-wrap.

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 1

    1. matteo-stratega/claude-cortex★ 8same 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 repository7 other files it gives its agents.

The file

Show its rules and text

We split this skill into the rules it gives an agent

30 rules. About 0 are common, 23 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈23

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Ask the user nine questions about their day and write their answers into today's daily note in their own words.

    line 8
  2. Distinctive:

    Start immediately. Do not explain what you are about to do, do not ask if they are ready, and do not open with a greeting.

    line 18
  3. Very specific:

    Read config/profile.md for the name, timezone, and life threads.

    line 23
  4. Distinctive:

    If not, the interview is nine questions.

    line 30
  5. Distinctive:

    Run these once, before question one, and use that date for everything the wrap writes.

    line 41
  6. Very specific:

    Before that first block, apply any taps waiting on the artifact pages: run "Applying taps" in artifacts/README.md.

    line 69
  7. Distinctive:

    Check the answer slots, not the file text.

    line 84
  8. Distinctive:

    Every slot still unfilled, nothing written at all: the user has not been set up.

    line 97
  9. Distinctive:

    Check whether a calendar connector is available.

    line 109
  10. Very specific:

    One at a time (default): Ask a question, wait for the answer, then ask the next.

    line 119
  11. Distinctive:

    If they skip ("nothing", "skip", "n/a"), move on and leave that section out of the note.

    line 134
  12. Distinctive:

    Do not analyze, reflect back, quote their answer approvingly, or offer an opinion mid interview.

    line 136
  13. Distinctive:

    Ask in this order:

    line 139
  14. Distinctive:

    Leave that section out, same as in the default mode.

    line 168
  15. Distinctive:

    If the reply is not numbered, has fewer answers than questions, or reads as combined or reordered, do not force a mapping.

    line 169
  16. Very specific:

    Only when config/season.md names non-negotiables.

    line 175
  17. Distinctive:

    Read back through the answers they just gave.

    line 179
  18. Distinctive:

    Never go item by item.

    line 228
  19. Distinctive:

    Never check a box the user did not tell you to check.

    line 230
  20. Distinctive:

    Never raise items that are due soon, or high priority with no due date.

    line 257
  21. Distinctive:

    If nothing is overdue, stale, or malformed, and nothing in the interview matched an open item, skip this step without comment.

    line 270
  22. Very specific:

    Keep that heading spelled exactly ## Decisions Made.

    line 319
  23. Distinctive:

    If the season file defines a custom question, its answer goes in a final subsection of the reflection, titled with a short phrase from the …

    line 323
  24. Distinctive:

    Write every answer verbatim.

    line 327
  25. Very specific:

    Only if config/season.md names non-negotiables.

    line 334
  26. Very specific:

    Run these as separate commands, never chained with &&.

    line 365
  27. Distinctive:

    Do not summarize the day back at them.

    line 386
  28. Distinctive:

    If they want to add something after the wrap is saved, treat it as a new capture rather than reopening the interview.

    line 390
  29. Distinctive:

    No coaching, no affirmations, no encouragement layer, during or after.

    line 398
  30. Distinctive:

    No em dashes and no double hyphens in anything you write.

    line 402

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Daily Wrap

Ask the user nine questions about their day and write their answers into today's daily note in their own words. This is a capture session, not a coaching session. Ask, listen, write, confirm.

When this runs

In the evening, on /daily-wrap or when the user says they are wrapping up. Once per day. If today's note already has an End of Day Reflection, say so and ask whether to add to it or replace it.

Start immediately. Do not explain what you are about to do, do not ask if they are ready, and do not open with a greeting. Ask question one.

Before you start

Read config/profile.md for the name, timezone, and life threads. Read config/season.md and note two things before the interview begins:

  • Non-negotiables (question 4 of the season file). If the user named any, you will write a status line for each at the end. If the answer is blank or (skipped), skip that section entirely and never mention it.
  • The custom evening question (question 5). If there is one, you ask it last. If not, the interview is nine questions.

Get the timezone out of config/profile.md and pass it to date through Bash, so the machine clock never decides what day it is. With Europe/Lisbon in the profile:

TZ=Europe/Lisbon date +%Y-%m-%d
TZ=Europe/Lisbon date +%A

Run these once, before question one, and use that date for everything the wrap writes. A wrap that starts at 11:50pm and ends after midnight still belongs to the day it started, so never re-run date partway through and never let one wrap write into two days' files.

Substitute whatever zone the profile names. Bare date +%A is wrong. Never infer a weekday from a date string. If the profile has no timezone in it, stop and ask for one before you write anything. Never guess a zone and never fall back to the machine clock, because a wrap that lands in yesterday's file looks correct and stays wrong.

The tracker check-in needs a date in the past, and past date arithmetic is not portable. On macOS the shell has BSD date and takes TZ=Europe/Lisbon date -v-14d +%Y-%m-%d. On Linux it has GNU date and takes TZ=Europe/Lisbon date -d '14 days ago' +%Y-%m-%d. Run one, and if the shell rejects the flag, run the other. Do not count back by hand.

Get it all in hand before question one

Nothing mechanical belongs between their last answer and the file getting written. Read everything the wrap needs up front, batched, so the interview and the check-in both run on data you already hold.

The timezone gates the date commands, so this comes in two rounds. First, in one block: config/profile.md, config/season.md, and tracker.md. Then, once the zone is known, in one more block: the three date commands above, today's calendar events if a connector is present, and a listing of meetings/.

Before that first block, apply any taps waiting on the artifact pages: run "Applying taps" in artifacts/README.md. The check-in reads the tracker, so the user's taps from today have to be in it first, or the check-in asks about items they already ticked off. When both URLs in config/artifacts.md are blank, or there is no ArtifactData tool in the session, this does nothing and you say nothing about it. The tracker edits it makes ride along in the wrap's own commit. Unmatched taps get one line each on the confirmation at the end, never in the middle of the interview.

Then ask question one. Do not read any of it a second time later. If one of those reads fails, carry on with the rest and say which part of the wrap ran short on the confirmation line, rather than dropping it in silence.

When the config is not fully filled

Check the answer slots, not the file text. Both config files carry instruction comments that describe the placeholder markers in prose, and those comments stay there forever, so a plain search for that shape matches on a fully configured vault every time. Look at what sits under each heading instead. Nothing in config/templates/ is config: those slots are filled by skills on every run, so never read them as setup answers.

A slot counts as filled when it holds anything the user put there, and (skipped) is something the user put there. Treat it as a filled answer, a deliberate pass on an optional question, never as a blocker.

Then degrade rather than stop:

  • Every slot still unfilled, nothing written at all: the user has not been set up. Say so in one line, offer the setup interview in CLAUDE.md, and stop.
  • Some filled, some not: run the wrap with what is there. No non-negotiables and no custom question means the interview is the nine questions and nothing else, with no mention of what is missing. No tag set means the note goes in without tags.

The one gap you cannot work around is the timezone, because the wrap has to know which day's file it is writing. Ask for it, wait, and do not guess.

A partly filled profile is a normal state, not an error.

Check whether a calendar connector is available. If one is, its read is part of that second block and you offer a pass over today's meetings as the interview closes. If not, the wrap runs exactly the same without it. Do not mention connectors you do not have.

The interview

The interview runs in one of two modes. Both ask the same questions and end in the same place.

  • One at a time (default). Ask a question, wait for the answer, then ask the next. This is what runs for a plain /daily-wrap.
  • Batch (on request only). Post every question in one message as a single numbered list, then parse the one reply. Use it only when the user asks for it: "batch mode," "all at once," "give me all the questions." If it is unclear whether they want batch, run one at a time.

Rules:

  • In the default mode, one question at a time. Wait for the answer before asking the next.
  • Warm and casual. This is a conversation, not a form.
  • Take short answers as complete. Push once only when a single word lands on a question that clearly wanted more, such as "fine" as the whole answer to how the day felt.
  • If they skip ("nothing", "skip", "n/a"), move on and leave that section out of the note. Do not write "nothing" into the file.
  • Do not analyze, reflect back, quote their answer approvingly, or offer an opinion mid interview. A brief acknowledgment, then the next question.

Ask in this order:

  1. "What was your biggest win today?"
  2. "What frustrated you most, or felt like a block?"
  3. "Did you pick up anything new? An insight, something that clicked, anything that surprised you?"
  4. "Any conversations or people worth remembering?"
  5. "Did you make any decisions today, even small ones?"
  6. "Anything unfinished or still on your mind that you don't want to lose?"
  7. "What's the one thing you want to carry into tomorrow?"
  8. "What are you grateful for today?"
  9. "Last one, how did the day feel overall? Energy, mood, anything."

Then, if config/season.md defines a custom evening question, ask it last, worded the way the user wrote it.

Question 5 does most of the work in this system. Small decisions count: what they said no to, what they picked between, what they stopped doing. If the answer is a single decision, that is fine. If they say none, leave the section out.

Parsing a batch reply

Batch mode only. The list you post includes the custom evening question, if there is one, as its last item. The non-negotiables sweep below is not in the list, because it depends on what they answer, so it comes after the parse.

  • Their numbered answers map by position to the list you posted: their "1." is your first question.
  • A missing number, or "skip," "n/a," or a bare dash against one, is a skip. Leave that section out, same as in the default mode.
  • If the reply is not numbered, has fewer answers than questions, or reads as combined or reordered, do not force a mapping. Ask one short follow-up that names the questions you cannot place, then carry on.

The non-negotiables sweep

Only when config/season.md names non-negotiables. This is the one question you are allowed to add, and you ask it after the last question, including the custom one.

Read back through the answers they just gave. Some non-negotiables will already be settled, because "got out for a walk at lunch" answers the walk. Take those as answered and do not ask again. Then ask about whatever is left, all of it in a single question:

"Two quick ones before I write this up: did you get outside today, and were you in bed by eleven?"

One question, however many items are unresolved. Never one question per item, and never ask about something they already covered. If everything is covered, ask nothing at all and say nothing about it. If they wave it off, that is fine too, write those lines as "not asked" and move on.

This is a checkbox sweep, not an opening. Take the yes or no, do not follow up, do not comment on the answer.

After the interview

Steps 1 and 2 are still the conversation. Do not announce a phase change, do not say you are about to go and run something, and do not pause, because the reads are already done and there is nothing to wait for. Keep going in the same voice and let the exchange end when the tracker check-in ends. Everything from step 3 down is silent.

1. Today's meetings, when a calendar connector is present

Today's events are already in hand. For any meeting that has ended and that has no note in meetings/, list them and ask once: "You had three meetings today with no notes. Want to capture anything from them?"

Take whatever they give you and route it through the capture skill, one meeting note per meeting. If they pass, move on and do not ask again.

With no calendar connector, skip this step silently.

This runs before the check-in on purpose. A note captured here becomes evidence the check-in reads a moment later, so do not reorder the two.

2. Tracker check-in

You read tracker.md before question one. Take items only from the thread sections below its ## Sections heading. Everything above that heading documents the format, and one line there has the shape of an unchecked item without being one. Skip any line holding a placeholder in curly braces, and skip anything inside a fenced code block wherever it sits, because a pasted example under a thread section reads as a real item and then surfaces as a phantom to-do every night until somebody notices it.

Surface at most five items total, grouped, one consolidated question per group. Never go item by item.

Never check a box the user did not tell you to check. A meeting note is evidence that a meeting happened, not evidence that their work around it is done, and a wrongly closed item disappears from every future briefing and review. So when a note in meetings/ is dated today and the item's own text names preparation or follow up for that specific meeting, say so and ask anyway, as part of the group's one question:

"Prep for the roof estimate, and there is a note from it, so I assume that one is done. Close it?"

Take the yes, then close it. If the item merely shares a person or a project with today's meeting, that is not evidence of anything, so treat it like any other item and ask plainly.

  1. Overdue (due: before today, unchecked): "These slipped past their due date. Done, pushed, or dropped?"
  2. Stale: two or three items with captured: fourteen or more days ago, no due:, not !high. "Still doing these?" Fourteen or more, not more than fourteen, so an item sitting at exactly fourteen days is caught here rather than falling between two definitions. The weekly recap draws the same line at the same number.
  3. Malformed: a line sitting under a thread section that reads like an open item and has no checkbox. Every rule above matches on - [ ], so an item written without one is invisible to this check-in, to the morning briefing, and to the weekly review, and it stays invisible for as long as it sits there. Say so once, offer to fix the syntax, and do not count it against the five.

Never raise items that are due soon, or high priority with no due date. Neither has slipped, and asking about future work at the end of the day is nagging. The morning briefing owns both groups. The check-in is for what already went wrong.

Apply their answer to tracker.md directly: check the box with a short outcome and the date, edit the text if the scope changed, add or move due:, change priority, or delete what they drop.

One more pass before you move on. If a win or a decision they described in the interview matches an open item, that item is done and they will not think to say so. Name it and ask: "Sounds like the permit call is done, close it?" This is where an undated to-do finally gets closed, so do not skip it.

If nothing is overdue, stale, or malformed, and nothing in the interview matched an open item, skip this step without comment.

3. Write into today's daily note

Find or create daily/YYYY-MM-DD.md from config/templates/daily-note.md, filling {{date}} and {{weekday}}. Fill the ## End of Day Reflection section:

## End of Day Reflection

### Biggest Win
[answer]

### Biggest Frustration
[answer]

### What I Learned
[answer]

### People Worth Remembering
[answer]

### Still On My Mind
[answer]

### Carry Into Tomorrow
[answer]

### Grateful For
[answer]

### How the Day Felt
[answer]

Then the decisions from question 5 go in their own section. ## Decisions Made is a top level section of the daily note and it sits directly before the ## End of Day Reflection section. Create it there if it is not there yet, and append to it if it is, because a capture earlier in the day may have opened it already and put it in that same place. Never open a second one.

## Decisions Made

- Dropped the Saturday market stall. Two months in and it never covered the gas.
- Told [[Dana Whitfield]] the roof work moves to April, rather than starting it in the rain.

Keep that heading spelled exactly ## Decisions Made. The weekly recap harvests the section by name, and the monthly and quarterly reviews are built on what the weekly finds. A renamed heading breaks the ladder quietly.

If the season file defines a custom question, its answer goes in a final subsection of the reflection, titled with a short phrase from the question itself.

Write every answer verbatim. Do not paraphrase, tidy the grammar, or upgrade their word choice. Leave out any section they skipped. While writing, wikilink every person and project mentioned, and add tags to the note's frontmatter from the set in config/profile.md.

4. Non-negotiables check

Only if config/season.md names non-negotiables. If it does not, this section does not exist and you never bring it up.

Append to today's daily note as the last section, below the reflection:

## Non-negotiables
- Walked outside: yes, got the dog out before the first appointment
- Asleep by 11: no, "up past one with the baby"
- Dinner at the table: not asked

Each line comes from something they said, either in the interview or in the sweep. Do not guess and do not infer a yes from silence. Use "not asked" when the sweep did not get to it or they waved it off, and it is better than a wrong yes. Record, do not advise. Three nights of "no" is the weekly recap's business, not yours.

5. Commit

Stage the files you actually touched, by path. Never git add -A, because it sweeps up whatever else is sitting in the folder:

git add daily/2026-03-14.md tracker.md
git commit -m "daily-wrap: 2026-03-14"

Run these as separate commands, never chained with &&. A commit with nothing to record exits non-zero, and a chain stops there without a word.

Then check for a remote. Push only to a remote the user owns, as its own command, whether or not the commit had anything new. If there is no remote, stop after the commit and say nothing about it.

Output

Three to five lines. Nothing else.

Wrap saved -> daily/2026-03-14.md
Decisions: 2 recorded
Tracker: closed the roof measurements and the permit call, pushed the gutter quote to 2026-03-20
Page taps: 2 applied

The page taps line appears only when the tap step applied or failed to match something. Name each unmatched one on its own line.

Do not summarize the day back at them. Do not tell them it sounds like a good day, a hard day, or any kind of day. Do not offer a suggestion for tomorrow. They just spent ten minutes telling you what happened, and they were there for it.

If they want to add something after the wrap is saved, treat it as a new capture rather than reopening the interview.

Tone and format rules

  • Their words stay theirs. "The whole day was a dumpster fire" does not become "the day had its setbacks." A year from now the value of this file is that it still sounds like them.
  • No coaching, no affirmations, no encouragement layer, during or after.
  • Skipped questions produce no section. A quiet day produces a short note.
  • Frontmatter stays at three fields: type, date, tags, tags as a quoted inline array.
  • No em dashes and no double hyphens in anything you write. Commas, parentheses, or a new sentence.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.