distill-my-life
Skill8 starsChanged 4 months agoMITView on GitHub
Distill each day's raw activity into a daily report. Ensures ~/.Life exists, pulls the day's activity via Luci, writes an objective report under ~/.Life/reflections/daily/, and writes new entities under ~/.Life/entities/. Use when the user says "distill", "distill my life", "catch up on daily reports", "summarize the last few days", or first needs the life folders created.
In this repository5 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- luci-profileSkill
- luciSkill
- summarize-meetingSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
30 rules. About 0 are common, 15 are distinctive, and 15 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈15
Distinctive · estimate
This author's own choices. Worth a look.
15
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 24
Create ~/.Life if it does not exist.
- Distinctive: line 36
Create any missing directory.
- Very specific: line 42
If ~/.Life/CLAUDE.md exists, use it.
- Distinctive: line 44
If neither exists, continue with the defaults in this skill (English, hourly blocks, no external send).
- Distinctive: line 48
If the index exists, follow its writing rules, daily-report format, and external connector config.
- Very specific: line 54
Read ~/.Life/RULES.md if it exists.
- Distinctive: line 57
Follow it for this whole run.
- Distinctive: line 59
If it says to leave a person, app, site, or subject out, that thing does not appear anywhere — not in a daily file, not in an entity file, …
- Very specific: line 62
Never write to
RULES.md: Do not create it, tidy it, reformat it, or append what you learned to it. - Very specific: line 70
Before writing anything, check whether ~/.Life/reflections/daily/ contains any file matching YYYY-MM-DD.md.
- Distinctive: line 78
Only complete days — yesterday and earlier.
- Distinctive: line 80
If the backlog exceeds 7 days, process only the most recent 7 and list the skipped dates explicitly at the end.
- Distinctive: line 116
Read the exit code of every command:
- Distinctive: line 127
Stop means: do not write a daily file for this day, do not touch later days, do not write entities from partial results, and do not fall …
- Very specific: line 137
Every daily file starts with this frontmatter.
- Very specific: line 167
Do not put a second copy of observation below the body.
- Distinctive: line 189
Never use a single timestamp without an end time.
- Very specific: line 196
Read ~/.Life/entities/ first.
- Very specific: line 224
If YAML is missing, insert name / kind frontmatter above the current body.
- Very specific: line 226
If today's date is not already a - YYYY-MM-DD: bullet, append one.
- Distinctive: line 227
Never edit existing bullets.
- Very specific: line 231
Only when ~/.Life/reflections/daily/ had no YYYY-MM-DD.md files before this run (Step 1) and Step 2 wrote at least one daily file, add one …
- Distinctive: line 241
If the file does not exist, skip this step — Luci has never run here.
- Distinctive: line 242
Read the whole file as JSON.
- Distinctive: line 260
Write the whole JSON object back to the same path, preserving every other key exactly as read.
- Very specific: line 268
After every daily file and entity is written, decide whether the user's profile, ~/.Life/profile.md, is due:
- Very specific: line 271
If the prompt that started this run has a ## Then write the profile section, or says the profile is not due, Luci has already decided.
- Very specific: line 280
When it is due, use the luci-profile skill to write it: a first profile when the file did not exist, the weekly rewrite otherwise.
- Distinctive: line 291
If a Luci command failed: the date it failed on, the CLI's message verbatim, and that no later day was processed.
- Very specific: line 297
If an external connector is configured in ~/.Life/CLAUDE.md or ~/.Life/AGENTS.md, send one summary message through it (exactly one per run).
≈ “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.
Files
<!-- luci-skill-version: 1.3.0 -->Distill — daily distillation
Luci's Life page reads these files as data. YAML and ## headings are
the schema. Prose under a heading is shown as written. Do not invent
fields the schema does not name.
Step 0: ensure the life file system
The life file system is always under the user's home directory, never the current repo:
- macOS / Linux:
~/.Life - Windows:
%USERPROFILE%\.Life
Create ~/.Life if it does not exist. Then ensure this layout:
~/.Life/
reflections/daily/ # daily reports (YYYY-MM-DD.md)
entities/
people/ # people
orgs/ # organizations
projects/ # projects
tools/ # software toolsCreate any missing directory. Do not ask first. Do not delete or move anything that is already there. Do not write into the current working directory.
Then look for a file-system index inside ~/.Life only:
- If
~/.Life/CLAUDE.mdexists, use it. - Else if
~/.Life/AGENTS.mdexists, use it. - If neither exists, continue with the defaults in this skill (English, hourly blocks, no external send). Do not create an index file — Luci writes both on first launch.
If the index exists, follow its writing rules, daily-report format, and external connector config. Missing keys fall back to the defaults above.
Step 0.5: read the user's rules
Read ~/.Life/RULES.md if it exists. It is written by hand by the user:
who is who, who and what to leave out, how reports should read.
- Follow it for this whole run. Where it disagrees with the index or with this skill, the rules file wins.
- If it says to leave a person, app, site, or subject out, that thing does not appear anywhere — not in a daily file, not in an entity file, not in the final output.
- Never write to
RULES.md. Do not create it, tidy it, reformat it, or append what you learned to it. It is the user's document; distill only reads it. - Missing file, or a file you cannot read: continue with the defaults. Do not create one.
Step 1: decide which days to process
- Before writing anything, check whether
~/.Life/reflections/daily/contains any file matchingYYYY-MM-DD.md. If it contains none, remember this — it means today's run may be the first distillation ever on this machine, which matters for Step 4. - Get today's date via Bash (never guess from memory).
- Range: from the last complete distillation to yesterday. A daily file
is complete only when its YAML frontmatter has both
dateandobservation. A file without those keys is missing — write it again. - Only complete days — yesterday and earlier. Never process today.
- Skip a date only when a complete file already exists.
- If the backlog exceeds 7 days, process only the most recent 7 and list the skipped dates explicitly at the end.
- A day with no data at all still gets a complete file: YAML plus one
observation, and no##capture blocks. Never fabricate. - "No data" means every Luci command for that day exited 0 and returned nothing. A command that failed is not a no-data day. See Step 2.
Step 2: write each daily file
Pull that day's activity via the Luci CLI, then write
~/.Life/reflections/daily/YYYY-MM-DD.md. Process days oldest first.
Pulling a day from Luci
Resolve <CLI> exactly as the luci skill describes (discovery file,
then PATH). Do not run status first: status never starts Luci, while
real commands start it in the background and wait for it. The first
usage call below is the preflight.
Ranges are epoch milliseconds in local time, <fromMs>:<toMs>. Get the
day's midnight from the shell, never from memory:
- macOS:
date -j -f '%Y-%m-%d %H:%M:%S' 'YYYY-MM-DD 00:00:00' +%sthen* 1000 - Linux:
date -d 'YYYY-MM-DD 00:00:00' +%sthen* 1000 - Windows:
[DateTimeOffset]::new([datetime]'YYYY-MM-DD').ToUnixTimeMilliseconds()
toMs is the next midnight minus 1.
- Screen:
<CLI> usage --tr <fromMs>:<toMs> --limit 500 --json. The limit is a hard cap and the CLI has no paging, so if exactly 500 entries come back, rerun per hour (24 ranges) and concatenate. - Audio:
<CLI> transcript "<keyword>" --tr <fromMs>:<toMs> --limit 500 --jsonfor the two or three subjects the screen entries show (meeting names, project names). There is no "all audio" command. - Details:
frame <ID>only for entries whose text is too thin to place.
Read the exit code of every command:
0with entries: use them.0with no entries: a real gap. Write the day as no-data if every command for that day was empty.3: your command was malformed. Fix it from<CLI> --helpand rerun. This is never the user's problem.2with "didn't respond in time" on stderr: rerun that one range once with a smaller--limitor per-hour ranges. If it fails again, stop.2with any other message, or1: stop.
Stop means: do not write a daily file for this day, do not touch later days, do not write entities from partial results, and do not fall back to guessing. Files already written for earlier days stay. Report the day it happened on and the CLI's stderr line verbatim (it is written for users and contains no paths). Then wait for the user to fix Luci and rerun. These failures are not expected when Luci is installed; do not loop on retries or try other tools.
YAML (required)
Every daily file starts with this frontmatter. observation is one
sentence. Slugs match entity filenames without .md. Use [] when a
list is empty. Always fold observation with > so colons in the
sentence cannot break the file:
---
date: 2026-08-24
samples: 187
audio_segments: 148
first: "01:21"
last: "23:06"
observation: >
Only 37 minutes of captured screen time before the 21-hour gap, and all
of it went into recording and re-recording the same talk.
projects:
- microsoft-luci-electron
people:
- matt-pocock
orgs: []
---samples is the number of screen entries usage returned for the day.
audio_segments is the number of distinct transcript segments you saw;
omit it when you ran no transcript query. first and last are the
local times of the earliest and latest screen entry. Omit a key only
when its source was not queried, never because a command failed (a
failed command means Step 2 stopped and this file is not written).
Do not put a second copy of observation below the body. The Life page reads
the YAML field.
Body
After the frontmatter:
# YYYY-MM-DD (Weekday)using the English weekday.- The screen-memory caveat in one short paragraph.
- Hourly blocks. Align screen activity and live transcription by timestamp. Objective only. If the source is uncertain, leave it out.
Heading grammar is locked. Use an en dash between times. Use an em dash
before a title or a gap label. No other ## shape:
## HH:MM–HH:MM
## HH:MM–HH:MM — title
## HH:MM–HH:MM — no samples ({duration})A gap is a heading only, no body. Never invent hours to fill a gap. Never use a single timestamp without an end time.
Step 3: write entities
While summarizing, watch for people, organizations, projects, and software tools.
Read ~/.Life/entities/ first. Write every entity that appeared in that
day's evidence — do not ask:
- people →
~/.Life/entities/people/<slug>.md - organizations →
~/.Life/entities/orgs/<slug>.md - projects →
~/.Life/entities/projects/<slug>.md - software tools →
~/.Life/entities/tools/<slug>.md
Slug: lowercase ASCII, digits, hyphens. Reuse the existing filename when
the name already has a file (match YAML name or the # heading).
New file:
---
name: Microsoft-Luci-Electron
kind: project
---
# Microsoft-Luci-Electron
- 2026-08-24: eight threads still in flight, including Insights unlockkind is person, org, project, or tool.
Existing file:
- If YAML is missing, insert
name/kindfrontmatter above the current body. Do not delete the body. - If today's date is not already a
- YYYY-MM-DD:bullet, append one. - Never edit existing bullets. Never rewrite prose into a summary.
Step 4: first-time notification
Only when ~/.Life/reflections/daily/ had no YYYY-MM-DD.md files
before this run (Step 1) and Step 2 wrote at least one daily file, add
one congratulations row to Luci's local notification center so it appears
next time the app opens. Do this at most once ever — skip it entirely if an
entry with id first-distill already exists.
- Find Luci's config file:
- macOS:
~/.luciMicrosoft/luci-config.json - Windows:
%USERPROFILE%\.luci\luci-config.json - Linux:
~/.luci/luci-config.json
- macOS:
- If the file does not exist, skip this step — Luci has never run here.
- Read the whole file as JSON. If it fails to parse, skip this step; never overwrite a config file you cannot read.
It has a
notificationskey holding an array (treat it as[]if absent). If any entry in that array already has"id": "first-distill", skip — already sent. Otherwise prepend this object to the front of the array, touching no other key in the file:{ "id": "first-distill", "kind": "distill", "title": "You just distilled your first day!", "body": "Your first daily report is in ~/.Life. Keep it up and Luci will build a real picture of your work over time.", "at": <current time as epoch milliseconds>, "read": false, "to": "/v2/life" }- Write the whole JSON object back to the same path, preserving every other key exactly as read.
This only makes the row appear the next time the app is opened or restarted — Luci does not watch this file live.
Step 5: the profile
After every daily file and entity is written, decide whether the user's
profile, ~/.Life/profile.md, is due:
- If the prompt that started this run has a
## Then write the profilesection, or says the profile is not due, Luci has already decided. Do what it says and skip the rest of this step. - Otherwise it is due when either:
~/.Life/profile.mddoes not exist and~/.Life/reflections/daily/holds at least 3 daily files, or- it exists and its frontmatter
updated:date is more than 7 days before today.
When it is due, use the luci-profile skill to write it: a first profile
when the file did not exist, the weekly rewrite otherwise. If you don't have
that skill, install it from https://github.com/Memories-ai-labs/Luci-skills
first. When it is not due, write nothing.
Final output
- Which folders were created in Step 0, if any.
- Which days got reports (list the file paths), including days rewritten because YAML was missing.
- Which dates were skipped and why.
- If a Luci command failed: the date it failed on, the CLI's message verbatim, and that no later day was processed.
- Which entity files were written or appended (if any).
- Whether
~/.Life/RULES.mdwas found and applied. - Whether the Step 4 first-time notification was sent.
- Whether Step 5 wrote the profile (first or weekly), or why it was not due.
- If an external connector is configured in
~/.Life/CLAUDE.mdor~/.Life/AGENTS.md, send one summary message through it (exactly one per run); if not configured, send nothing.
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.

