start-day

name	start-day
description	Use when starting the day to set up today's daily note and process any prior unprocessed notes.
allowed-tools	Bash, Read, Write, Edit, Glob, Grep, AskUserQuestion, Agent, mcp__claude_ai_Google_Calendar__list_events
model	claude-sonnet-4-6

Morning startup for an Obsidian vault using the PARA structure. Runs in two phases: first closes out any unprocessed daily notes from prior days (routing content to its destination and archiving), then primes today's note with rolled-over todos, Avoidance Radar items, and context. Non-interactive except for ambiguous 2_Areas folder matches.

Vault resolution — at the top of the run, determine which vault is active:

obsidian vaults

• If ObsidianPersonal appears, set $VAULT=ObsidianPersonal — this is the personal vault mode (Daily Spark, Life Domains, Quotes, inspiration all apply).
• Otherwise (e.g. ObsidianWork), set $VAULT to that vault name — this is non-personal vault mode (skip Daily Spark, skip Life Domains domain nudge, skip the templated Daily Note write — fall back to a simpler obsidian daily + append flow). Avoidance Radar still applies if the file exists.

All vault=ObsidianPersonal references below should be read as vault=$VAULT — substitute the resolved vault name throughout. Sections marked [personal-vault only] are skipped in non-personal mode.

Key rules:

• Vault parameter comes immediately after the subcommand: obsidian <subcommand> vault=$VAULT [options]
• Use path= for exact vault-relative paths; file= for wikilink-style name resolution
• Never run obsidian --help — it hangs and never exits

---

Phase 1 — Close out unprocessed notes

Step 1a — Find unprocessed daily notes and fire bootstrap reads

Fire all of the following in one parallel batch:

# Today's note path — extracts inbox folder and today's date
obsidian daily:path vault=ObsidianPersonal

# VAULT_PATH — needed for find commands below; strip the leading "=> " prefix
obsidian eval vault=ObsidianPersonal code="app.vault.adapter.basePath"

# Phase 2 bootstrap reads — fire now so results accumulate while Phase 1 runs
mcp__claude_ai_Google_Calendar__list_events(
  calendarId="[e-mail]",
  startTime="<today>T00:00:00",
  endTime="<today>T23:59:59",
  orderBy="startTime",
  timeZone="America/Los_Angeles"
)
obsidian read vault=ObsidianPersonal path="2_Areas/Quotes.md"
obsidian search vault=ObsidianPersonal query="tag:#inspiration" format=json
obsidian read vault=ObsidianPersonal path="2_Areas/Life Domains.md"
obsidian search vault=ObsidianPersonal query="[agent-context:vault]" format=json

# Project canonical-file modification times for staleness check (vault API).
# Convention: a project's canonical file is named after its folder, e.g.
# `1_Projects/Foo Project/Foo Project.md`. (The legacy `Index.md` naming is no
# longer used — wikilinking to a verbose project name is more readable than to `[[Index]]`.)
obsidian eval vault=ObsidianPersonal code="JSON.stringify(app.vault.getFiles().filter(f => f.path.startsWith('1_Projects/') && f.parent && f.basename === f.parent.name).map(f => ({path: f.path, mtime: f.stat.mtime})).sort((a,b) => a.mtime - b.mtime))"

# Today's weather forecast — requires $WEATHER_LAT_LONG and $WEATHER_TZ env vars
curl "https://api.open-meteo.com/v1/forecast?${WEATHER_LAT_LONG}&daily=temperature_2m_max,temperature_2m_min,precipitation_probability_max&current=weather_code&${WEATHER_TZ}&past_days=0&forecast_days=7&wind_speed_unit=mph&temperature_unit=fahrenheit&precipitation_unit=inch" | jq -r '{high_temperature: .daily.temperature_2m_max, low_temperature: .daily.temperature_2m_min, precipitation_chance: .daily.precipitation_probability_max} | map_values(max) | "Todays Forecast: High \(.high_temperature)°F, Low \(.low_temperature)°F, with a \(.precipitation_chance)% chance of precipitation"'

After VAULT_PATH resolves, fire this second batch (also in parallel):

# Inbox contents (Phase 1 input — folder derived from daily:path result)
obsidian files vault=ObsidianPersonal folder="0_Inbox"

# Weekly note for current ISO week
CURRENT_WEEK=$(date +%Y-%V)
obsidian files vault=ObsidianPersonal folder="4_Archive/Weekly Notes"

# Project-scope agent-context files modified within 14 days
find "$VAULT_PATH/1_Projects" -name "*.md" -mtime -14 | xargs grep -l "^agent-context: project" 2>/dev/null

Store all results under these labels for use in Phase 2: CALENDAR_EVENTS, VAULT_PATH, QUOTES_RAW, INSPIRATION_FILES, LIFE_DOMAINS, VAULT_AGENT_CONTEXT, INDEX_MTIMES, WEEKLY_FILES, PROJECT_CONTEXT_FILES, WEATHER_FORECAST. If any call fails, store empty/null and continue silently.

Filter the inbox list to files matching the YYYY-MM-DD.md pattern where the date is before today. Sort ascending (oldest first). These are unprocessed prior notes.

If none exist, skip to Phase 2.

Step 1b — Process each prior note (oldest first, no pausing between notes)

The only exception to no-pausing is an ambiguous 2_Areas folder match (see Step 1b-iii/iv).

Track for each note: count of todos, avoidance items, idea stubs, learnings filed. Carry extracted action items forward in memory — they feed directly into Phase 2.

Multiple prior notes (2+): Dispatch one Agent subagent per note (model: haiku) to run Steps 1b-i through 1b-iv concurrently. Each subagent creates stubs and files learnings directly. However, each subagent returns avoidance items as output rather than writing them — the main thread does a single Avoidance Radar read + batch append after all subagents finish (avoids write conflicts on a shared file). After all subagents complete, the main thread runs Step 1b-v (distillation + archive) for each note in date order.

Step 1b-i — Read, resolve wikilinks, and classify

obsidian read vault=ObsidianPersonal path="0_Inbox/YYYY-MM-DD.md"

If the note is empty or contains only a template skeleton with no substantive content, skip to Step 1b-v (archiving).

Wikilink batch search — extract ALL [[...]] references from the note text at once. Launch all searches simultaneously in one parallel batch (both folders for every link at the same time):

obsidian search vault=ObsidianPersonal folder="1_Projects" query="<link1>" format=json
obsidian search vault=ObsidianPersonal folder="2_Areas" query="<link1>" format=json
obsidian search vault=ObsidianPersonal folder="1_Projects" query="<link2>" format=json
obsidian search vault=ObsidianPersonal folder="2_Areas" query="<link2>" format=json
# … all links at once, not one at a time

If a match is found in either folder and is not already in the project context map, read the matched file and any sibling files with agent-context: project (or agent-context: area if under 2_Areas/) in their frontmatter, then add them to the map. This surfaces projects referenced intentionally but outside the 14-day auto-load window.

Classify every meaningful item into one of five types (most specific match wins):

Type	What it looks like
Hard todo	An explicit next action: "I need to", "call X", "book Y", "follow up on Z"
Avoidance item	Something being put off — reappears across notes, reluctant tone
Reflection	Mood, mental state, day observations — not actionable
Learning	A fact, concept, or insight acquired from reading, a call, or experience
Idea fragment	A half-formed thought with no clear next step yet

Items under ## Today's Priorities are hard todos by default (unchecked only — skip - [x]).

Project attribution — for each hard todo, fuzzy-match its text against project names in the project context map. A direct name mention, a wikilink to the project, or a domain clearly owned by one project all count. Tag each todo internally with its matched project, or mark it "unattributed" if no match.

Step 1b-ii/iii/iv — Create stubs, log avoidance, file learnings (parallel)

Run two batches in sequence:

Read batch (fire together):

obsidian read vault=ObsidianPersonal path="2_Areas/Personal Knowledge Management/Avoidance Radar.md"
obsidian files vault=ObsidianPersonal folder="2_Areas"

Write batch (fire together after reads complete):

All three write types target different files — fire them simultaneously.

Idea stubs — for each idea fragment:

obsidian create vault=ObsidianPersonal path="0_Inbox/Idea - <Short Title>.md" content="---\ntags:\n  - idea\n  - stub\nsource: \"[[YYYY-MM-DD]]\"\n---\n\n<raw thought, verbatim or lightly cleaned>" silent

Keep <Short Title> to 4–6 words. Don't expand or elaborate.

Avoidance items — fuzzy-match each item against the Radar content read above. If not already present, append:

obsidian append vault=ObsidianPersonal path="2_Areas/Personal Knowledge Management/Avoidance Radar.md" content="\n- [ ] <item> — *first noted: YYYY-MM-DD*"

If the Radar file doesn't exist, create it first:

obsidian create vault=ObsidianPersonal path="2_Areas/Personal Knowledge Management/Avoidance Radar.md" content="---\nagent-context: vault\n---\n\nThings that keep coming up but keep getting deferred. Reviewed weekly.\n" silent

When running as a subagent (multi-note dispatch): skip writing to the Radar entirely — return avoidance items as output to the main thread instead.

Learnings — match each learning semantically to a 2_Areas/ subfolder (from the files list). Clear match: proceed without asking. No match: suggest top 2 candidates or offer to create a new folder — this is the only pause point. Once answered, continue:

# Append to existing note
obsidian append vault=ObsidianPersonal file="<Note Name>" content="\n---\n\n<learning content>\n\nSource: [[YYYY-MM-DD]]"

# Or create a new note
obsidian create vault=ObsidianPersonal path="2_Areas/<Subfolder>/<Title>.md" content="---\ntags:\n  - <tag>\n---\n\n<learning content>\n\nSource: [[YYYY-MM-DD]]" silent

Step 1b-v — Write Distillation, navigation footer, archive

Distillation — append before archiving:

obsidian append vault=ObsidianPersonal path="0_Inbox/YYYY-MM-DD.md" content="## Distillation\n\n**Day in brief:** <1–2 sentence synthesis of the day's tone, main themes, and energy — drawn from project context where applicable, not just the surface text>\n\n**Action items:**\n\n*[[Project Name]]:*\n- [ ] <item> *(<brief context informed by project state>)*\n\n*Unattributed:*\n- [ ] <item> *(<brief context>)*\n\n**Avoiding** *(added to Avoidance Radar):*\n- <item, or omit block if none>\n\n**Ideas stubbed to Inbox:**\n- [[Idea - <title>]]\n\n**Filed to 2_Areas:**\n- [[<Area Note>]] ← <learning summary>"

Omit any subsection with nothing to show. For Action items: group todos under their attributed project as a subheading. If all todos are unattributed, skip the grouping and list flat. If only one project is represented, skip the project subheading and annotate each item inline with *(→ [[Project Name]])*. Omit the Unattributed heading if everything is attributed.

Navigation footer — the note may already have a ← [[PREV-DATE]] line (added by Phase 2 when today's note was primed). If so, replace it with the full bidirectional footer using a direct file edit on the full filesystem path ($VAULT_PATH/0_Inbox/YYYY-MM-DD.md):

perl -i -pe 's|\Q← [[PREV-DATE]]\E|← [[PREV-DATE]] \| [[NEXT-DATE]] →|' "$VAULT_PATH/0_Inbox/YYYY-MM-DD.md"

If no footer line exists yet (note was never primed by Phase 2), append instead:

obsidian append vault=ObsidianPersonal path="0_Inbox/YYYY-MM-DD.md" content="\n\n---\n← [[PREV-DATE]] | [[NEXT-DATE]] →"

• PREV-DATE: calendar day before this note's date
• NEXT-DATE: calendar day after this note's date

Archive:

obsidian move vault=ObsidianPersonal path="0_Inbox/YYYY-MM-DD.md" to="4_Archive/Daily Notes/YYYY-MM/YYYY-MM-DD.md"

---

Phase 2 — Prime today's note

Step 2a — Create today's note

obsidian daily vault=ObsidianPersonal

Creates today's note from the template if it doesn't exist.

Step 2b — Load context

Bootstrap batch results from Step 1a should already be available. If any are still pending, wait for them now before continuing.

Read vault agent-context files — for each path returned in VAULT_AGENT_CONTEXT, read the file. These provide who-I-am, life-domain, and long-running-goals context.

Read project-scope agent-context files — for each path in PROJECT_CONTEXT_FILES, read the file. Build the project context map: project name (from path or frontmatter title) → key details (goal, current status, constraints, blockers). This map is shared across Phase 1 and Phase 2.

Weekly note — from WEEKLY_FILES, find the file matching $CURRENT_WEEK.md. If found, read it.

• Extract ## Weekly Focus — if the section exists, take the single sentence beneath the heading and store as WEEKLY_FOCUS_CONTENT (the verbatim sentence, no heading, no surrounding whitespace). If the section is absent, the file is empty, or the weekly note doesn't exist at all, leave WEEKLY_FOCUS_CONTENT empty. This is the only content from the weekly note that gets replicated into today's daily note — it powers the <!-- weekly-focus --> placeholder.
• Everything else is AMBIENT CONTEXT ONLY — do not replicate other sections into today's daily note. It informs internal calibration: which rolled-over todos feel weightier (project matched ## Projects Focused), which open ## Decisions Needed items deserve attention, whether to flag a stale project (already covered by ## Archive Candidates).

Avoidance Radar — read now (after Phase 1 has finished updating it):

obsidian read vault=ObsidianPersonal path="2_Areas/Personal Knowledge Management/Avoidance Radar.md"

Rolled-over todos:

• If Phase 1 processed any notes: use the action items already in memory from those Distillations — skip the archive re-read.
• If Phase 1 had nothing to process: list archived notes and read the most recent one to extract rolled-over todos:

obsidian files vault=ObsidianPersonal folder="4_Archive/Daily Notes"
obsidian read vault=ObsidianPersonal path="4_Archive/Daily Notes/YYYY-MM/YYYY-MM-DD.md"

Find ## Distillation → **Action items:**, extract every - [ ] line verbatim.

Recurring todo escalation — before building priorities, check for todos stuck across 3+ days. Read the two archived Distillations immediately preceding the source note and extract their unchecked - [ ] action items. Any todo text that fuzzy-matches a rolled-over todo in both prior notes has been deferred for 3+ consecutive days — escalate it: append to Avoidance Radar (Step 1b-iii format, first noted: = today's date) and remove it from rolled-over todos. Skip this check if fewer than 2 prior archived Distillations exist.

Step 2b-ii — Pick today's quote and inspiration [personal-vault only]

Skip this entire step in non-personal vault mode. Leave QUOTE_CONTENT, INSPIRATION_PATH, and INSPIRATION_TEASER empty/null.

Quote — from QUOTES_RAW, extract every line beginning with - . Strip the leading - . Use today's date as a seed: take the day-of-year (1–366), mod by the number of quotes, and select that line. Same quote all day if the skill runs twice.

Store as QUOTE_CONTENT — the raw line text including attribution and any [[wikilink]].

Inspiration — from INSPIRATION_FILES, apply the same day-of-year mod against the list to pick one file deterministically. Read only the first non-frontmatter, non-heading line of that file (skip --- blocks and lines starting with #) — stop after the first prose paragraph; do not load the full file.

Store as INSPIRATION_PATH (vault-relative path) and INSPIRATION_TEASER (the first prose line).

Step 2c — Build PRIORITIES_CONTENT and CONTEXT_CONTENT

Rolled-over todos — first items in PRIORITIES_CONTENT, verbatim from Distillation.

Avoidance Radar items — for each - [ ] line, extract item text and first noted: date. Calculate days elapsed from today.

Age	Action	Format
< 3 days	Skip	—
3–6 days	Bottom of list	- [ ] Item
7–13 days	Mid-list	- [ ] Item *(N days)*
14–20 days	Near top	- [ ] Item *(N days — still waiting)*
21+ days	Top of list	- [ ] Item *(N days — decide or drop it)*

Sort oldest first. Append after rolled-over todos.

Stale project flag — from INDEX_MTIMES (the canonical project files, e.g. 1_Projects/Foo/Foo.md), if any is 30+ days old, pick the least stale one and add:

- [ ] [[Project name]]: no activity in N days — worth a push?

Skip if a rolled-over todo already references that project. Skip if everything is under 30 days.

Cap — if PRIORITIES_CONTENT exceeds 6 items, remove from the bottom: youngest Radar items first, then stale flag. Never drop rolled-over todos.

CONTEXT_CONTENT:

• Line 1 (personal-vault only): One sentence from Life Domains ## Current Context — the single most time-sensitive or seasonally relevant point. Skip in non-personal mode.
• Line 2 (conditional): Only if any Radar item is 14+ days old: Overdue: [item name] (N days), [item name] (N days). Omit entirely if nothing qualifies.
• Line 3 (conditional): Only if CALENDAR_EVENTS is non-empty: **Meetings:** HH:MM Event title, HH:MM Event title — list each event's start time (12-hour, no seconds) and summary, comma-separated. Omit entirely if no events.

Step 2d — Write to today's note

Get today's note path:

obsidian daily:path vault=ObsidianPersonal

Full filesystem path: $VAULT_PATH/ + returned path (reuse VAULT_PATH from Step 1a).

In non-personal vault mode, skip the Python template-rewrite block entirely. Instead, run obsidian daily vault=$VAULT to create today's note from whatever default template the vault has, then append two sections: ## Rolled-over Todos (PRIORITIES_CONTENT) and ## Morning Context (CONTEXT_CONTENT, omitting any empty lines). Append the yesterday-nav footer the same way.

In personal vault mode, use a single Python script to do all substitutions and write the final file. Always read from the template file directly — do not patch the obsidian daily-created file, which has a known CLI bug that double-encodes multibyte characters (emoji, °, curly quotes) and will corrupt the note.

Template path: $VAULT_PATH/3_Resources/Obsidian Templates/Daily Note Template.md

INSPIRATION_FILENAME is the bare note name (no path, no .md) for the wikilink. The *italics* wrapper around the quote placeholder is already in the template. If WEATHER_FORECAST is empty/null, substitute an empty string for <!-- weather -->.

WEEKLY_FOCUS_LINE — derive from WEEKLY_FOCUS_CONTENT (extracted in Step 2b):

• If WEEKLY_FOCUS_CONTENT is non-empty: WEEKLY_FOCUS_LINE = f"{WEEKLY_FOCUS_CONTENT} — [[{CURRENT_WEEK}]]" — this puts the focus sentence followed by a wikilink to the weekly note for one-click access.
• If WEEKLY_FOCUS_CONTENT is empty (no weekly note, or no ## Weekly Focus section in it): substitute an empty string. The surrounding callout still has quote/inspiration/weather, so the missing line collapses gracefully.

python3 << 'PYEOF'
vault = "$VAULT_PATH"
template_path = f"{vault}/3_Resources/Obsidian Templates/Daily Note Template.md"
note_path = f"{vault}/<TODAY_VAULT_PATH>"

with open(template_path, 'r', encoding='utf-8') as f:
    content = f.read()

content = content.replace('<!-- weekly-focus -->', 'WEEKLY_FOCUS_LINE')
content = content.replace('<!-- quote -->', 'QUOTE_CONTENT')
content = content.replace('<!-- inspiration -->', '[[INSPIRATION_FILENAME]] — INSPIRATION_TEASER')
content = content.replace('<!-- weather -->', 'WEATHER_FORECAST')
content = content.replace('<!-- priorities -->', 'PRIORITIES_CONTENT')
content = content.replace('<!-- context -->', 'CONTEXT_CONTENT')
content += '\n\n---\n← [[YESTERDAY-DATE]]'

with open(note_path, 'w', encoding='utf-8') as f:
    f.write(content)
PYEOF

Embed all content values directly into the heredoc before running — do not use shell variable interpolation for the replacement strings, as they may contain characters that break shell quoting. If the note already contains a nav footer line (starts with ←), the skill already ran today — skip the write entirely.

---

Phase 3 — Summary

If Phase 1 processed any notes, show the processing table:

Date	Todos	Avoidance	Idea stubs	Learnings filed	Archived
2026-04-10	2	—	—	—	✓

One row per note. Show — for columns with nothing found. If a note was empty and skipped processing, note that in the row.

If Phase 1 had nothing to process, no summary needed — the note is ready silently.