| name | reflect |
| description | Daily reflection -- what was done, depth of work, analysis. Triggers: /reflect, reflection, daily review, how was my day, what did I do today. |
Daily Reflection
A meaningful daily reflection -- not just a task list, but analysis of what happened,
the volume of work behind it, what went well and what didn't.
Principle: Two data layers -- task delta (what moved) + work narrative (what actually happened).
Without the second layer, reflections will be dry and incomplete.
Chain: /reflect (day) -> /retro (week) -> /monthly-review (month)
Step 1: Get Today's Date
get_today() -- MANDATORY first call, for date filtering
Remember today's date in YYYY-MM-DD format (for filtering) and DD.MM (for display).
Step 2: Gather Task Delta (Layer 1)
Read from all your task sources -- with a backward filter: what CHANGED/CLOSED today.
CUSTOMIZE: Replace these with your actual task sources.
Parallel batch 1 (task manager + calendar):
trello_list_cards(list_name="done") -- what was closed
trello_list_cards(list_name="today") -- what's still in today (planned but not done)
list_calendar_events() -- today's events
Parallel batch 2 (vault sources):
read_vault("work/project/tasks/TASKS.md") -- completed items with today's date
read_vault("dashboard.md") -- items with today's date
Backward filter logic:
| Source | What to look for in reflection |
|---|
| Task manager done | Cards/tasks with activity date = today. Done list may be cumulative -- ALWAYS filter by date! |
| Task manager today | What's left = what was planned but NOT done |
| Work tasks | Rows marked done with today's date |
| Dashboard | Active tasks with today's date. Completed with today's date |
| Calendar | Events on today's date (meetings, deadlines) |
Step 3: Gather Work Narrative (Layer 2)
Session notes = depth of work. Task list shows WHAT, session notes show HOW and HOW MUCH.
Read session index:
read_vault("conversations/CONVERSATIONS.md")
Filter by today's date:
Find all entries in CONVERSATIONS.md containing today's date.
Extract paths to session files.
Read each session file:
For each session from today -- read via read_vault().
Extract:
- "What was done" -- concrete actions
- "Decisions made" -- decisions taken
- Tags -- domains (project-a, brain, job-search, content, etc.)
IMPORTANT: Read session notes in parallel (up to 5 at a time).
Step 4: Analyze & Synthesize (Layer 3)
4.1 Group by domain
Combine data from Layer 1 (tasks) and Layer 2 (sessions) into domains:
CUSTOMIZE: Define your domains based on your life/work areas.
- Work Project -- work tasks + calendar events + sessions tagged with project
- Job Search -- job search changes + sessions tagged job-search
- Server/Brain -- dashboard items + sessions tagged brain/server
- Content -- content plan + content tasks + sessions tagged content
- Personal -- personal tasks (gym, chores, etc.)
- Other -- everything that doesn't fit above
4.2 For each domain:
- What was done -- tasks + depth from session notes
- Key decisions -- from "Decisions made" in sessions
- Work volume -- time labels from task manager + number of sessions
4.3 Meta-analysis:
- Plan vs reality -- what was in today list this morning vs what actually closed
- Unplanned -- tasks that appeared during the day (not in morning plan)
- Missed -- what's still in today / fail items / overdue
- Balance -- which domains dominated, what didn't get attention
- Energy -- hours (by task labels), session count, intensity
Step 5: Present Reflection
Machine-readable time header (universal):
When saving the reflection to vault, prepend a YAML front-matter block with time-tracking
fields. Source the numbers from the invisible <message-meta> tag injected on each user
message by the inject-message-timestamp hook (now / session-start / elapsed / episodes /
gap) -- work_minutes is the SUM of episode durations (a 45-min idle gap splits episodes),
NOT wall-clock between first and last message.
---
session_start: HH:MM
session_end: HH:MM
work_minutes: N # sum of episode durations, not wall-clock
episodes: N # count of 45-min-bounded work bursts
tags: retro, daily
---
Format:
Reflection for DD.MM (day of week)
OVERVIEW
-- Tasks closed: N (from task manager) + N (from work/dashboard)
-- Work sessions: N
-- Estimated time: ~X.X hours (by task labels)
-- Dominant domain: [domain]
[DOMAIN 1]
[Tasks + depth from sessions. NOT just a list -- describe what actually happened]
[DOMAIN 2]
[What happened, who you talked to, what decisions were made]
...
WINS
-- Top achievements of the day (2-3 items)
MISSED / DIDN'T FIT
-- What's left + why (conscious choice or lack of time?)
KEY DECISIONS
-- Decisions from sessions that affect the future
TOMORROW
-- What's already in today list (from Layer 1)
-- Leftovers from today (overdue, fail)
-- Recommendation (what's higher priority, what to focus on)
Style:
- Not a dry list -- narrative, as if talking to a person
- Depth -- for each domain, not "task closed" but what was behind it
- Honesty -- if the day was overloaded, say it. If something was missed, don't hide it
- Specifics -- numbers, names, decisions. Not "worked on project", but "had a call with Alex, reviewed 5 screens, assigned 4 tasks"
Step 6 (optional): Tomorrow Preview
If user asks about tomorrow ("what's tomorrow?", "plans for tomorrow"):
trello_list_cards(list_name="today") -- current today list = tomorrow's plan
list_calendar_events() -- tomorrow's events
read_vault("work/project/tasks/TASKS.md") -- deadlines for tomorrow
Show: what's already planned + what needs to be added (leftovers, overdue).
Important Rules
- get_today() FIRST -- always, no exceptions
- Session notes = mandatory layer -- without them, reflection will be superficial
- Task manager done = key source, but filter by date! -- done list may be cumulative. Each card has a date. Take ONLY cards with today's date
- User can narrow scope -- "only work tasks", "only after 3pm" -> filter accordingly
- Ask user if clarifications needed (which tasks are today's, context on fail items)
- Discussion-first -- a reflection (or any significant conclusion / assessment) is a draft for discussion, not a silent auto-report. Present it, let the user react/correct, THEN fixate. Don't bury a debatable judgement inside a file the user never read
- AUTO-SAVE -- after presenting reflection AND incorporating any corrections, save to vault:
write_vault("retro/daily/YYYY-MM-DD.md", ...) with the time-YAML header + tags retro,daily + day's domains
- Chain with /retro -- daily reflection feeds weekly retro. If end of week, suggest "/retro?"
Related Skills
/overview -- forward-looking task aggregator (what to do). Reflect = backward-looking (what was done)/retro -- weekly retrospective. Uses archived done list for week stats/monthly-review -- monthly aggregation of retros/track -- task routing (write). Reflect is read-only analysis
