ワンクリックでManusで任意のスキルを実行

process-journal

スター0

フォーク0

更新日2026年6月5日 04:17

Process a morning journal entry — extract AI summary, mood, priorities, gratitude, and people mentioned. Full insights go to the journal entry; all priorities grouped as Must do / Focus work / If time go to the daily hub. Extracted priorities become Todoist tasks (with dedup and batch approval). Use when the user says "process journal", "process today's journal", or "/process-journal".

インストール

Codex または Claude でインストールこの Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。

Manusで実行

ソース

RS42-AI

RS42-AI/ai-os-lite

GitHub リポジトリを開く Creator のリポジトリを見る

ダウンロード

Manusで実行

Process Journal

Extract structured insights from a raw morning journal transcript. Write the full AI analysis to the journal entry (private) and a grouped priority list to the daily hub (visible).

Key Principles

The raw transcript stays untouched. The voice transcript under ## Morning and ## Evening is never modified.
Full insights go to the journal entry. The ### AI Summary section (with mood, all priorities, gratitude) replaces the placeholder at the bottom of the journal entry (nested under ## Morning). This is the private reflection space.
The daily hub gets Today's priorities. This is the ONLY place priorities are written — not the morning entry. ALL priorities are included (work and personal), grouped by urgency: Must do / Focus work / If time. The grouping replaces the old area-based filter.
No silent failures. Track every external source call. Report what succeeded, what failed, and what was skipped. See vault-config/references/source-manifest.md.

Sources

Track all source calls per vault-config/references/source-manifest.md. This skill uses:

Source	Used For	Criticality
Obsidian CLI	Read journal transcript, write AI Summary	REQUIRED
Obsidian MCP	Patch sections, create task notes	REQUIRED
QMD Search	People lookup in vault contacts	LOW
Todoist CLI	Task creation, dedup check	HIGH

Invocation

/process-journal                       → process today's entry
/process-journal 2026-03-08            → process a specific date
/process-journal 2026-05-18 --dry-run  → render to stdout/tmpfile, do not write the vault

Dry-run mode (--dry-run flag):

Step 6 (AI Summary) and Step 8 (daily hub render) write their would-be content to /tmp/process-journal-dryrun-$today.md instead of patching the vault.
Step 7 (task notes / Todoist) is skipped entirely — no creates, no Todoist API calls.
Step 4 (idempotency marker check) still runs, but a detected marker just emits a warning — no prompt to re-run.
The bottom marker append (Step 7g) is skipped.

Pass --dry-run as the second argument after any date: /process-journal 2026-05-18 --dry-run.

Workflow

Tool Rules

Reference: vault-config/references/tool-selection.md CLI for reads, writes, graph traversal, and property operations. MCP only for semantic search and section-level patching. DO NOT use mcp__obsidian-mcp-tools__get_vault_file for reads.

Step 1: Determine Target Date

Default: today's date
If argument provided: parse as YYYY-MM-DD
Validate format

Step 2: Read Source Files

Use Obsidian CLI (NOT MCP) for reads:

# Journal entry (raw transcript)
obsidian read path="5. Resources/Personal/Journal/Morning Entries/YYYY-MM-DD.md"

# Daily note hub (target for Today's priorities)
obsidian read path="1. Daily/YYYY-MM-DD.md"

DO NOT use mcp__obsidian-mcp-tools__get_vault_file — use obsidian read per the Tool Rules above.

If journal entry doesn't exist: Report "No journal entry found for YYYY-MM-DD" and stop.

If daily note doesn't exist: Report "No daily note found for YYYY-MM-DD. Create one first with Cmd+Shift+D in Obsidian." and stop.

Step 3: Read Habit Data from Frontmatter

Read all habit_* properties present in the journal entry's frontmatter — do not assume a fixed set. The habit schema is owned by the journal template (system-settings/Templates/Journal Entry Template.md); this skill reads whatever habit_* fields the entry actually carries. They are the single source of truth — the user sets them via Obsidian's Properties panel.

Extract every habit_* line from the frontmatter block:

obsidian read path="5. Resources/Personal/Journal/Morning Entries/YYYY-MM-DD.md" \
  | awk '/^---$/{c++; next} c==1 && /^habit_/'

This lists each habit_*: value pair the entry carries (currently habit_workout, habit_meditation, habit_gratitude, habit_vitamins — but read what's there, don't hardcode). habit_gratitude may be set to true during Step 6 if 3 gratitude items are extracted.

No parsing or syncing needed — frontmatter IS the data.

Store the values for use in the Step 9 report (habits summary).

Step 4: Check Idempotency — Bottom Marker Line

Scan the journal entry body for a bottom marker line matching: *Processed YYYYMMDDHHMM · N tasks created in Todoist*

Search command:

obsidian read path="5. Resources/Personal/Journal/Morning Entries/YYYY-MM-DD.md" | grep -E '^\*Processed [0-9]{12}'

If no marker found: this is a first run — proceed normally. Step 7 will append a marker at the end.

If a marker is found: this is a re-run. Extract the timestamp and Todoist count from the marker. Warn the user:

This journal was already processed on YYYY-MM-DD at HH:MM ({N} Todoist tasks created).
Re-processing will refresh the AI Summary, mood, and grouped priorities.
Task notes and Todoist tasks that were already created will be preserved (not duplicated).

Would you like to re-process? [y/N]

Wait for confirmation. If the user declines, stop with a short message.

On re-run (user confirmed): extract the existing priorities from the current ### AI Summary and the daily hub's **Today's priorities:** section before re-extracting in Step 5. These will be merged (dedup by description) in Step 8.

Frontmatter flag: the old todoist_tasks_created frontmatter flag is no longer used. If present on an old entry, ignore it — the marker line is the sole idempotency source.

Step 5: Extract Insights from Journal

Extraction is no longer transcript-only. It runs five reads in parallel and merges the outputs into a unified "candidate priorities" list with provenance. Each candidate carries {text, source, project_hint, lane_hint, context, ai_capability} so Step 7 (task-note creation) and Step 8 (render) can set the right frontmatter tags and capability markers respectively.

5a: Read all five sources

#	Source	How
1	Today's transcript (`## Morning` section)	`obsidian read path="5. Resources/Personal/Journal/Morning Entries/$today.md"` then extract under `## Morning`
2	Prior-day hub	`bash ${CLAUDE_PLUGIN_ROOT}/skills/process-journal/scripts/parse_prior_hub.sh $yesterday`
3	`/start-day`'s in-journal Warm/Cold recap	`bash ${CLAUDE_PLUGIN_ROOT}/skills/process-journal/scripts/parse_startday_recap.sh $today`
4	Project hub `Current Status` snapshots (only for projects named in transcript)	`obsidian read path="<hub>"` then locate `## Current Status`
5	Contact context for each named person	`bash ${CLAUDE_PLUGIN_ROOT}/skills/process-journal/scripts/resolve_contact.sh "<name>"` per name

Run sources 2, 3, and 5 unconditionally. Run source 4 only for projects the transcript explicitly names. Source 1 is the spine.

5b: Carry-forward reconciliation (against prior-day hub)

For every bullet in source #2 (parse_prior_hub.sh output):

If the bullet has a wikilink_target that resolves to a task note via read_task_status.sh:
- status == "done" → drop. The work shipped; do not carry forward.
- status in {"todo", "in-progress", "on-hold"} → carry as a candidate with lane_hint: "carry-forward".
- status empty / found: false → carry as a candidate with lane_hint: "carry-forward" and a (no task note) marker.
If the bullet had no wikilink at all (plain text): carry as a candidate with (no task note) and lane_hint: "carry-forward".
Annotate each carried bullet with its prior lane in the context field (e.g. "was p1 due TODAY"). Step 8's render uses this for the (carry-forward from <date>) marker on the absorbing lane.
Note (5/18 delta — Decision 1): carry-forward items do NOT render in a separate "Yesterday's open items" lane. They get absorbed into Must do / Focus work / Could-hand-off-to-AI per their content, with a (carry-forward from <date>) marker. Items with no concrete today-action go to "Also on the radar" with (carry-forward p2 from <date>; no concrete edge today — stay aware) per Decision 9.

Caveat on blocked_by parsing: read_task_status.sh reads blocked_by as a scalar via fm_value. When the task's blocked_by: frontmatter is an empty list [], the script returns the literal string "[]". Treat both "" and "[]" as "not blocked" in 5d below.

5c: Warm-band carry-through (against `/start-day` recap)

For each entry in source #3 (parse_startday_recap.sh output):

If the project's slug appears in the transcript (case-insensitive substring match against transcript text): no carry-through; the transcript-driven extraction (5a #1) will surface it directly.
If the project's slug does not appear in the transcript:
- If a today's transcript-extracted priority resolves to a parent area or project that this project is a child of (e.g. {Project2} under "Re-orient on {WorkArea}"), emit as a sub-bullet candidate with lane_hint: "sub-project-under-<parent>".
- Otherwise emit as a standalone candidate with lane_hint: "also-on-radar" and context: "{band}, {days_silent}d".
Cold band entries follow the same logic but always default to also-on-radar (no sub-project promotion).

Parent-resolution heuristic: a warm project is a "child" of a transcript priority when (a) they share an area, AND (b) the transcript priority's text contains a phrase like "re-orient on", "catch up on", "audit", or "review" referencing the area — the kind of meta-priority that naturally umbrellas sub-projects. When uncertain, render as standalone "Also on the radar".

5d: Blocker inference (propose-confirm-record loop) — 5/18 delta extends this

For each candidate priority surfaced by 5a/5b/5c that has an associated project slug:

Read the task note's blocked_by frontmatter via read_task_status.sh.
- Filled with wikilinks → use as-is. Skip inference. The chain renders inline on the bullet per Decision 3 of the 5/18 delta.
- Literal none → user has confirmed no blockers. Skip inference. Render normally.
- Empty / missing / "[]" / no task note → run inference (next step).
Run the blocker scanner:
```
bash ${CLAUDE_PLUGIN_ROOT}/skills/process-journal/scripts/scan_blockers.sh "<priority text>" "<project slug>" "<transcript path>"
```
~10 second budget per priority. If the scanner returns zero hits, move on (no prompt). If hits are very high (>20), summarize the top 5 by source diversity before presenting.

If the scanner returns hits, propose them in a single prompt to the user:

[[Priority]] — possible blockers I found:
  • <description> (source: <source_path> · <source_section>)
  • <description> (source: <source_path> · <source_section>)

[Confirm + record]  [Edit]  [Skip — not blocked]

On Confirm:
- For each proposed blocker that resolves to an existing task note → use its wikilink.
- For each proposed blocker without a task note → flag for Task-Note Creation in Step 7 (do not auto-create here; the user reviews the batch then).
- 5/18 delta (Decision 6) — Insight→action rule: For EACH confirmed chain, ALSO propose the first concrete unblock-step task (verb + target). Examples: "Email NACA advisor to set meeting date", "Confirm IRS extension status and schedule tax-return prep block". The user confirms creation; Step 7 (the task-creation step) then creates the task note with ← unblocks <chain> reasoning written into the body. No chain gets confirmed without producing at least one concrete actionable task — vague execution-strategy steps are not acceptable.
- Write blocked_by: to the priority's task note frontmatter (only if the task note exists).
On Edit: ask which blockers should actually apply. Re-prompt with the corrected list. Same Decision-6 rule on confirmation.
On Skip: write blocked_by: none to the task note. This is the permanent "user confirmed no blockers" marker — never re-infer.

If no task note exists for the priority and the user confirms blockers, the blocked_by write is deferred to Step 7 (the task-creation step, when the task note is created). The chain back-reference renders inline on the bullet in Step 8 (the render step) from the in-memory result, formatted per the 5/18 delta:

← unblocks <plain description>
← part of the <cluster name> (e.g. "the reply-debt cluster")
← starts the <path name> (e.g. "the NACA approval path (next: advisor meeting → approval → city decision)")

5e: Contact re-grounding (per named person)

For each first-name mentioned in the transcript:

Run bash ${CLAUDE_PLUGIN_ROOT}/skills/process-journal/scripts/resolve_contact.sh "<name>".
If found: true: use the canonical wikilink + context summary in any priority bullet that mentions this person.
If found: false: use the raw name as a red wikilink ([[Raw Name]]) so clicking it creates the contact page later.
For names that appear in the AI Summary's People mentioned line (Step 6), use the same resolved wikilinks. Do not duplicate the resolution.

5f: Extract from transcript (the original Step 5 logic, narrowed)

After 5b/5c/5d/5e produce their candidate sets, extract from the transcript itself:

AI Summary (2–3 sentences) — same as v1. Concise summary of what the person discussed, their state of mind, and key themes. Write in third-person observational tone.
Mood — same as v1. Single word or short phrase capturing the emotional tone. Examples: "reflective, motivated", "stressed but hopeful", "energized", "contemplative".
Themes on the mind (new) — recurring topics that aren't actionable today. Pattern: "I wonder if…", "I've been thinking about…", "we keep coming back to…". Render in the AI Summary in Step 6.
Today's transcript-explicit priorities — verbs of intent ("I want to", "I need to", "we have to", "let's", "going to"). Each becomes a candidate with source: "transcript" and a project_hint extracted from surrounding context (project name mentioned, area implied).
Research questions (new) — patterns like "I wonder if X can…", "can we…", "is there a way to…". 5/18 delta (Decision 2): these become first-class task-note candidates with tags: [task, research]. They do NOT render in a separate "Research questions" lane — they surface via the existing Tasks Overview Base on the daily hub. Treat as a regular candidate with lane_hint determined by Section 8a rules; the research tag is added in Step 7 (task-note creation).
AI-handoff candidates (new) — explicit language like "AI can definitely handle", "the AI should be able to", "dispatch this". Each becomes a candidate with lane_hint: "could-hand-off-to-ai" and ai_capability: "ai-handoff" (set in 5h).
Gratitude — same as v1. Must be explicitly stated; never inferred. If not stated, mark missing: **Gratitude**: ⚠️ No gratitude entry today. What are 3 things you're grateful for?
People mentioned — feed into 5e.

5g: Merge candidates with dedup

Combine the candidate lists from 5b + 5c + 5f into one list. Dedup by (project_slug, normalized_text) — two candidates that point to the same underlying work merge, with the union of their context fields kept on the surviving entry. Provenance is preserved (source: ["transcript", "carry-forward"] etc.) for the Step 8 parenthetical suffix.

When a transcript candidate matches a carry-forward candidate (same task note or same normalized text): the carry-forward annotation is dropped from the parenthetical (the transcript reinforces today; no need to surface "carry-forward" since it's already actively on the mind). The (carry-forward from <date>) marker only renders when the user did NOT re-mention it.

5h: AI capability classification (5/18 delta — Decisions 4–5)

For each merged candidate from 5g, classify into one of three AI-capability levels. This drives the task-note tags: frontmatter (Step 7, the task-creation step) and the bullet capability marker (Step 8, the render step).

Classification	Tag set	Bullet marker	Rule
`human-only`	`[task]`	(unmarked)	Replies, person-to-person comms, identity-level decisions, anything requiring tacit context the AI can't fully see
`ai-handoff`	`[task, ai-handoff]`	🤖	Codebase work + research + structured queries with clear inputs — AI can run autonomously now
`ai-pending-decision`	`[task, ai-pending-decision]`	⏳	Code/work that depends on a specific human decision named in the note body

Persist the classification on the candidate as ai_capability. Step 7 (task-creation) reads it to set tags: on the task note; Step 8 (render) reads it to render the marker prefix.

Note on GWS-CLI deferral: Reply-drafting via GWS-CLI is deferred until permissions land. Once shipped, draftable replies graduate to ai-pending-decision (AI drafts → user reviews → user sends). Until then, replies stay human-only.

Step 6: Write AI Summary to Journal Entry

The ### AI Summary heading already exists at the bottom of the journal entry (placed by ensure_journal.sh with placeholder text, nested under ## Morning). Replace its content using a heading-targeted patch.

mcp__obsidian-mcp-tools__patch_vault_file(
    filename="5. Resources/Personal/Journal/Morning Entries/YYYY-MM-DD.md",
    operation="replace",
    targetType="heading",
    target="AI Summary",
    content="""**Summary**: [extracted summary]
**Mood**: [extracted mood]
**Gratitude**:
1. [item 1]
2. [item 2]
3. [item 3]
**People mentioned**: [[Person 1]], [[Person 2]]
"""
)

What's NOT in the AI Summary:

Habits — already tracked in frontmatter properties (single source of truth). Weekly review aggregates from frontmatter.
Priorities extracted — actionable ones flow to the daily hub (Step 8, the render step) and are classified for task note creation (Step 7). No need to duplicate in AI Summary.

People mentioned — always use [[wikilinks]] for people, even if no contact page exists yet (red links in Obsidian prompt page creation later). Search vault contacts first for full names:

mcp__qmd__search(query="{Contact}", collection="vault")

If found: [[{Contact} {ContactLastName}]] (full name from contact page) If not found: [[{Contact}]] (red link — creates page when clicked)

Re-processing (merge behavior): If the user re-processes an already-processed entry (Step 3 confirmed):

Summary and mood: Replace with fresh extraction (these are subjective — always use the latest read)
Gratitude and people: Replace with fresh extraction
Today's priorities: merge existing + newly extracted priorities in Step 8i. Deduplicate by comparing description text. Preserve checkbox state on existing items.

Use replace on the existing AI Summary heading instead of append. Do NOT include ### AI Summary in the content when using replace — the heading itself is preserved by the patch tool.

Step 6b: Task Note Cross-Referencing

Cross-reference extracted priorities against existing task notes:

For each extracted priority, use the shared helper:
```
bash ${CLAUDE_PLUGIN_ROOT}/skills/start-day/scripts/match_task_note.sh "distinctive keywords"
```
This searches all */Tasks/*.md locations (2. Projects/{Area}/Tasks/, 2. Projects/{Area}/{Project}/Tasks/, Personal/Tasks/, Personal/{Project}/Tasks/).
If a matching task note exists: note it for the daily hub display (include task note status and priority)
If no task note exists and the priority warrants one (specific action, due date, multi-session): flag for creation in Step 7 (task-creation step)

Step 6c: Previous Evening Chain Link (safety net)

/start-day Step 4b renders the Previous Evening link inside the ### Last Night's Reflection blockquote. This step is a safety net for when /start-day was skipped: if the ### Last Night's Reflection section has no Previous Evening link, insert one inside that section's blockquote:

> **Previous Evening**: [[5. Resources/Personal/Journal/Evening Entries/YYYY-MM-DD-1|Last Night's Evening]]

Where YYYY-MM-DD-1 is yesterday's date. If the link is already present (the normal case — /start-day ran), do nothing. Never insert it at the top of the file.

Step 7: Create Task Notes and Todoist Mirrors

Task notes are the source of truth. Todoist is a notification/reminder layer. This step creates task notes first, then optionally mirrors them to Todoist. This step runs BEFORE Step 8 (render) so that every actionable bullet on the rendered hub wikilinks to a real task note — no (no task note — create) placeholders.

7a: Classify each priority

Classification	Criteria	Action
Task note warranted	Specific action, has a due date, spans sessions, or p1/p2	Create task note (7c), optionally mirror to Todoist (7e)
Quick action	Sub-15-min, no tracking needed	Todoist-only (7e) or skip entirely
Reflection	Vague intention, life decision, habit/routine	Skip — leave in journal only

7b: Dedup against existing task notes

For each priority classified as "task note warranted", run:

bash ${CLAUDE_PLUGIN_ROOT}/skills/start-day/scripts/match_task_note.sh "distinctive keywords"

If a matching task note is found:

Do NOT create a new task note
If the matched task note has status: done, note it for the batch presentation ("mentioned again — reopen?")
Otherwise, use the existing task note (its wikilink will be placed in Step 8)

7c: Route new task notes to the correct folder

Per spec Section 4 (Area-Level Task Routing):

Priority's area	Task note location
Work/business project ({WorkArea}/{SideArea2}/{SideArea} + project)	`2. Projects/{Area}/{Project}/Tasks/{Task Name}.md`
Work/business area (no project)	`2. Projects/{Area}/Tasks/{Task Name}.md`
Personal life project (APAC-Trip, Health, Finances, etc.)	`Personal/{Project}/Tasks/{Task Name}.md`
Personal life area (no project)	`Personal/Tasks/{Task Name}.md`

Never route task notes to the knowledge-notes folder — task notes live exclusively under */Tasks/ per the table above.

7d: Task note frontmatter

The tags: block is set from the candidate's ai_capability field (assigned in Step 5h):

`ai_capability`	`tags:`
`human-only` (default)	`[task]`
`ai-handoff`	`[task, ai-handoff]`
`ai-pending-decision`	`[task, ai-pending-decision]`

If the candidate is a research question (per Step 5f), the research tag is added on top of the above. These tags mirror to Todoist/Linear labels via Step 7e (the Todoist mirror sub-step).

# Determine tags based on the candidate's ai_capability (from Step 5h)
if candidate.ai_capability == "ai-handoff":
    tags = "tags:\n  - task\n  - ai-handoff"
elif candidate.ai_capability == "ai-pending-decision":
    tags = "tags:\n  - task\n  - ai-pending-decision"
else:  # human-only (default)
    tags = "tags:\n  - task"

# If candidate is a research question (per Step 5f), add `research` tag
if candidate.is_research_question:
    tags += "\n  - research"

mcp__obsidian-mcp-tools__create_vault_file(
    filename="{routed_path}",
    content=f"""---
date: YYYY-MM-DD
type: task
status: todo
area: {area_slug}
project: {project_slug}
priority: {p_level}
due_date: "{YYYY-MM-DD if known else empty}"
scheduled_date: ""
done_date: ""
blocked_by: ""
external_id: ""
{tags}
---

# {Task Name}

> **Project**: [[{Project Hub}]]

{1-2 sentences of context from the journal transcript}

## Dev Log

<triple-backtick>base
filters:
  and:
    - type == "devlog"
    - tasks == "[[{Task Name}]]"
views:
  - type: table
    name: Related Sessions
    order:
      - date
      - file.name
      - session_topic
    sort:
      - property: date
        direction: DESC
<triple-backtick>
"""
)

(Replace <triple-backtick> above with three literal backticks when writing actual task note content. The skill doc can't inline nested triple-backticks cleanly.)

7e: Optional Todoist mirror

For each new task note (or existing task note re-mentioned today), decide whether to mirror to Todoist:

Create Todoist mirror?	When
Yes	Has a due date, or priority p1/p2, or user needs a reminder notification
No	Reference task that lives in the task note alone (e.g., "figure out X")

If mirroring:

td task add "Task description" \
  --project "{AreaProject}" \
  --labels "{area_label}" \
  --priority p3 \
  --due "today"

Record the Todoist task ID in the task note's external_id frontmatter:

obsidian property:set path="{task note path}" property="external_id" value="Todoist: {task_id}"

7f: Present batch for approval

Before creating anything, present the batch:

Task proposals from YYYY-MM-DD journal:

CREATE TASK NOTES ({N}):
├─ [{Area}/{Project}/Tasks/{Name}] — {summary} [priority, due_date]
└─ ...

MIRROR TO TODOIST ({M}):
├─ Task note {Name} → Todoist "{Name}" [project, p-level, due]
└─ ...

TODOIST-ONLY / QUICK ACTIONS ({Q}):
├─ "{text}" → Todoist [project, p4]
└─ ...

SKIP ({S} — not actionable):
├─ "{text}" — reflection
└─ ...

EXISTING TASK NOTES ({E} — already present):
├─ {Name} (status: todo) — mentioned again, comment added to Todoist
└─ ...

Create these? [Y/n/edit]

Wait for user confirmation. If user says "edit", ask which items to modify. If "n", skip creation and proceed to Step 7g.

7g: Append bottom marker line

After all creation steps complete (even if the user skipped all), append a marker line to the journal entry body.

First run (no existing marker):

obsidian append path="5. Resources/Personal/Journal/Morning Entries/YYYY-MM-DD.md" content="
*Processed ${timestamp} · ${todoist_count} tasks created in Todoist*"

Where timestamp is YYYYMMDDHHMM (e.g., 202604201239) and todoist_count is the number of Todoist tasks actually created in this run.

Re-run (existing marker detected in Step 4): REPLACE the existing marker line in place — do NOT append a second marker line. Use direct file edit (read the file, sed-replace the *Processed ... line, write back). Appending would compound the MCP patch_vault_file bottom-append bug that motivated this refinement.

Do NOT set any frontmatter flag — todoist_tasks_created and journal_processed are both dropped. The marker line is the sole idempotency source.

Step 8: Write Today's Priorities to Daily Hub

This step composes the 6-lane priorities block and writes it to the daily hub via a single MCP heading patch on Morning Journal. The lanes are populated from the merged candidate list produced in Step 5g, classified by lane_hint, then rendered with capability markers, inline chain back-references, and contextual suffixes. By the time this step runs, Step 7 has already created task notes for every actionable candidate — so every bullet wikilinks to a real note, no (no task note — create) placeholders.

8-pre: Read existing hub state

obsidian read path="1. Daily/$today.md"

Confirm the ## Morning Journal heading exists at H2. On a re-run, extract any existing bullets under each lane for merge (8i).

8a: Lane classification — assign each candidate to one of six lanes

Walk each candidate from Step 5g; assign to exactly one lane based on the rules below (first-match wins, top to bottom):

Lane	Triggers (any match → this lane)
Must do	priority p1 or p2 (from task note); OR due today/overdue; OR transcript uses urgency language ("I have to", "I need to today", "overdue", "must"); OR `lane_hint == "reply-debt"` (set when context contains "promised", "owed", "follow-up rule")
🤖 Could hand off to AI	`ai_capability == "ai-handoff"` (set by 5h) AND not Must-do urgency
Focus work	the priority has a project hint matching today's session work, OR carries a parent for sub-project promotion (5c step 2), OR is the transcript's primary topic
If time	low-energy language ("maybe", "eventually", "if I get to it", "would be nice"); OR no urgency signal and no project anchor
Also on the radar	`lane_hint == "also-on-radar"` (set by 5c); OR `lane_hint == "carry-forward"` AND the candidate has no concrete today-action (Decision 9)
Execution strategy	synthesis lane — populated in 8e after all bullet lanes are settled

Carry-forward absorption rule (5/18 delta — Decision 1): candidates with lane_hint: "carry-forward" that DO have a concrete today-action are absorbed into Must do / Focus / Could-hand-off-to-AI per the rules above, with a (carry-forward from <date>) marker appended to their context. Carry-forwards WITHOUT a concrete today-action go to "Also on the radar" with (carry-forward p2 from <date>; no concrete edge today — stay aware).

Research-question candidates (tags: [task, research] from Step 5f / Step 7 task-note creation) classify like any other candidate — they don't get a dedicated lane. They surface separately via the existing Tasks Overview Base on the daily hub.

If a candidate matches none of the lanes above, default to Focus work.

8b: Meta-theme subtitle synthesis

Synthesize a single (meta-theme: …) subtitle from the transcript + carry-forward state. Goal: one phrase that captures what the day is about. Examples from prior days:

*(meta-theme: system-build has been blocking deliverables; today is about closing the loop — ship process-journal v2, batch the comms debt, dispatch the autonomous-handoff layer)*

Heuristics:

If a single project dominates Must do + Focus work, the meta-theme is "ship X" or "close the loop on X".
If carry-forward is heavy, the meta-theme references catching up or unblocking.
If reply-debt is heavy, the meta-theme references comms cleanup.

Render this subtitle on the priorities header (see 8g).

8c: Notation legend

Immediately under the priorities header, render this exact italic legend line:

> **Notation:** 🤖 = AI can run autonomously now · ⏳ = AI can run after the named human decision · *(unmarked)* = human-only · `← unblocks <X>` = back-reference to a chain context inline on bullets

8d: Render lanes 1–5

Render each lane only if it has at least one bullet. Lane order is FIXED:

Must do — with an optional subtitle when all bullets share a theme (e.g. *(reply debt — all overdue)*). Bullet format:
```
- [<marker?>] [[Task Note Wikilink]] *(<context>)* ← <chain back-reference if any>
- [<marker?>] [[Reply to Person]] *(48hr rule, 5+ wk past)* ← part of the reply-debt cluster
```
Marker 🤖 / ⏳ / (unmarked) per the task note's ai_capability from 5h. ALL named entities in the bullet must be wikilinks (Decision 12 — see Section 8f).

Focus work — with optional subtitle *(today's deep work — <synthesized theme>)*. Bullet format:

- [<marker?>] [[Task Note]] *(<carry-forward state if any>)* ← <chain back-reference>
- [[Re-orient on {WorkArea} work — status and next steps]] *(carry-forward p2 from 5/16; no concrete edge today — stay aware)*
  ↳ Warm sub-projects under this:
    • [[{Project3}]] — 7d since [[Request {Project3} license from {Contact1}]]
    • [[{Project2}]] — 14d since [[2026-05-04 - Post-PTO Re-entry, Goals Sent, Trainings Completed]] *(just flipped Warm → Cold)*

(Note: the carry-forward-p2-with-no-edge example actually belongs in "Also on the radar" per Decision 9 — included here only to show the sub-project nesting pattern. Sub-project bullets come from Step 5c's parent-resolution.)

If time — bullet format:

- [<marker?>] [[Task Note]] *(<context>)*

🤖 Could hand off to AI (dispatch in parallel — autonomous TODAY) — note the 🤖 prefix on the lane header itself as a section signpost. Bullet format:
```
- 🤖 [[Mobile app bug review and fix pass]] *([[Portfolio]] · cold, 71d)* — verbatim: "right now the mobile is kind of buggy..." ← first live test of the autonomous-dev pattern
```
Include a verbatim: "<transcript quote>" suffix when the transcript explicitly says "AI can handle" / "dispatch" / similar.

Also on the radar (transcript-named or implied, not today-actionable). Bullet format:

- [[Project Hub]] *(warm, 7d)* — <context for why mentioned>
- [[Carry-Forward p2 Item]] *(carry-forward p2 from 5/16; no concrete edge today — stay aware)*
  ↳ Warm sub-projects under this:
    • [[Sub-Project]] — Nd since [[recent file]]

8e: Synthesize the Execution strategy lane

Compose a numbered day plan that orders the Must do + Focus work + Could-hand-off-to-AI bullets into a feasible sequence. Every numbered step → points at a wikilink to a bullet already rendered above (Decision 8). No item lives only in the strategy — if a step describes work that has no corresponding bullet above, the bullet is missing from a lane and must be added.

Step format:

N. **<Phase label>** → <wikilink(s) to bullets above> (<lane reference>)

Examples (5/18 fixture):

1. **Now** → [[Ship /process-journal v2 today]] (Must do)
2. **Quick discovery** → [[Email NACA advisor to set meeting date]] + [[Confirm IRS extension status and schedule tax-return prep block]] (Must do)
3. **Reply batch — single 30-min block** → [[Reply to {Contact}...|{Contact}]] · [[Reply to {Contact1}...|{Contact1}]] · ... (Must do)
4. **🤖 Dispatch AI handoffs in background** → [[Mobile app bug review and fix pass|Portfolio fix]] · ... (Could hand off to AI — all N)
N. **Before EOD** *(housekeeping — not bullets)* → confirm carry-forward outcomes, lock fork decisions
N+1. **EOD** *(housekeeping — not bullets)* → `/vault-commit` · [[5. Resources/Personal/Journal/Evening Entries/$today|evening journal]]

Heuristic phase ordering:

Now — foreground session work (usually inferred from the most-recent transcript paragraph).
Quick discovery — small lookups / status checks that unblock later work.
Reply batch — batched reply debt.
🤖 Dispatch AI handoffs in background — Could-hand-off-to-AI lane in parallel.
Foreground (pick one) — the top Focus work item.
Before EOD (housekeeping — not bullets) — verification / anchor commitments.
EOD (housekeeping — not bullets) — /vault-commit, evening journal.

Omit any numbered step whose source lane is empty. Renumber sequentially.

8f: Wikilink discipline (Decision 12)

Every named entity in the rendered hub MUST be a wikilink to its canonical vault note:

People → 4. Contacts/People/<Name>.md (resolved via 5e contact re-grounding; voice-fuzz tolerance handled by resolve_contact.sh)
Actions → task notes in */Tasks/ (created by Step 7 — the task-creation step that ran before this render)
Projects → project hubs in 2. Projects/<Area>/<Project>/ (used only when a project-level reference is meant, not a specific action)
Devlogs, design notes, plans → their canonical paths
Housekeeping references ({SideArea} hubs, {Project1}, {Project2} in /vault-commit context) → project hubs

Pipe aliases ([[Long Task Name|short label]]) are encouraged where the canonical filename is verbose.

8g: Compose the full block + patch

Assemble in this exact order (omit empty lanes):

> [[5. Resources/Personal/Journal/Morning Entries/$today|Open Morning Entry]]

**Today's priorities** *(meta-theme: <synthesized day theme from 8b>)*

> **Notation:** 🤖 = AI can run autonomously now · ⏳ = AI can run after the named human decision · *(unmarked)* = human-only · `← unblocks <X>` = back-reference to a chain context inline on bullets

**Must do** *(<subtitle if unifying theme>)*:
- [<marker?>] [[Task Note]] *(<context>)* ← <chain back-reference>
- ...

**Focus work** *(today's deep work — <theme>)*:
- ...

**If time** *(<subtitle if fork decision>)*:
- ...

**🤖 Could hand off to AI** *(dispatch in parallel — autonomous TODAY)*:
- 🤖 [[...]] — verbatim: "..." ← <chain ref>
- ...

**Also on the radar** *(transcript-named or implied, not today-actionable)*:
- [[Project Hub]] *(warm, Nd)* — <context>
- ...

**Execution strategy** *(today's order — every actionable bullet above maps to a step here)*:
1. **Now** → [[...]] (<lane>)
2. **<Phase>** → [[...]] (<lane>)
...
N. **Before EOD** *(housekeeping — not bullets)* → ...
N+1. **EOD** *(housekeeping — not bullets)* → ...

Two trailing blank lines after the last lane, then the existing --- separator stays intact.

Patch via:

mcp__obsidian-mcp-tools__patch_vault_file(
    filename=f"1. Daily/{today}.md",
    operation="replace",
    targetType="heading",
    target="Morning Journal",
    content="<full block above>",
)

8h: Guard against duplicate sections

After patching, re-read the daily hub. If a duplicate Morning Journal section appears at the bottom (known MCP patch_vault_file issue), remove it via direct file edit.

8i: Re-run merge behavior

If Step 4's idempotency check detected an existing bottom marker and the user confirmed re-run:

Re-classify all candidates fresh (today's date may have shifted urgency).
For each lane, dedup new bullets against existing bullets in the same lane (compare by normalized description text after stripping wikilinks).
Preserve any ← unblocks <X> chain annotations from existing bullets unless inference found different chains (then prompt: "Replace existing chain context? [y/N]").
Re-render the entire block; do NOT attempt surgical bullet-level patches.

Step 9: Report Results

Display a summary followed by the execution report (per vault-config/references/source-manifest.md):

Journal processed for YYYY-MM-DD.

**Summary**: [2-3 sentence summary]
**Mood**: [mood]
**Habits**: N/M completed (list checked items, note missed items)
**Priorities**: [count] items extracted, grouped to the daily hub as:
  - Must do: N
  - Focus work: N
  - If time: N
**Todoist**: [created count] tasks created, [initiative count] initiatives, [skip count] skipped, [existing count] already existed
**Gratitude**: [3 items written to journal / already present / ⚠️ not mentioned — prompt shown]
**People**: [list or "none mentioned"]

Journal entry updated: [[5. Resources/Personal/Journal/Morning Entries/YYYY-MM-DD]]
Daily hub updated: [[1. Daily/YYYY-MM-DD]]

---
### Execution Report
#### Sources
- [x] Obsidian CLI — journal read, daily hub read/write
- [x] Obsidian MCP — N sections patched
- [x] QMD Search — people lookup, N contacts resolved
- [x] Todoist — N tasks created, M deduped

#### Warnings
- [only if there are actual warnings]

#### Fix
- [only if there are failed sources with actionable fixes]

Only include the Warnings and Fix sections if there are actual warnings.

Verification (5/18 fixture)

The 2026-05-18 daily hub is the byte-level regression target for the v2 render shape. After any change to Step 5 / Step 7 / Step 8, run a dry-run against 2026-05-18 and diff vs the committed hub.

Acceptance criteria

Acceptable deviations (paraphrasing):

Lane subtitle wording (e.g. "reply-debt + chain-unblockers" vs "reply debt — all overdue")
Chain descriptor wording inside ← unblocks <X> annotations
Execution-strategy step labels (the phase labels, not the underlying wikilinks)
Meta-theme subtitle wording
The exact ordering of bullets within a single lane (provided the lane content is equivalent)

NOT acceptable (regressions — investigate and fix):

Lane order differs from: Must do · Focus work · If time · 🤖 Could hand off to AI · Also on the radar · Execution strategy
Any lane present in the committed hub is missing in the rendered output (or vice versa)
Capability markers (🤖 / ⏳ / (unmarked)) on bullets missing or wrong
Any actionable bullet rendering as plain text instead of a wikilink to a task note — i.e. (no task note — create) placeholders appearing post-Step-7
Inline ← unblocks <X> / ← part of the <cluster> / ← starts the <path> annotations missing on chain bullets
The > **Notation:** ... legend line missing under the priorities header
The *(meta-theme: ...)* subtitle on the priorities header missing
Sub-project nesting under "Also on the radar" carry-forward-p2 items (↳ Warm sub-projects under this:) missing
Step 7 (task creation) running AFTER Step 8 (render) — sequence flip is mandatory per Decision 7

Regression triage

If a "not acceptable" item appears, trace back:

Missing lane or wrong ordering → Step 8a lane classification rules
Missing capability marker → Step 5h AI capability classification OR Step 7d task-note tag wiring
(no task note — create) placeholder → Step 7 (task creation) failed to run before Step 8 (render); check workflow ordering
Missing ← unblocks annotation → Step 5d blocker inference loop didn't propose the chain, or 8d/8e didn't render it
Missing meta-theme subtitle → Step 8b synthesis didn't run or returned empty
Missing notation legend → Step 8c literal-string render missing

Live-fire test

The first real verification is the user's morning invocation: /process-journal on today's transcript. Confirm all 6 lanes render (or are correctly omitted when empty), carry-forward absorbs prior open items, blocker inference asks the right questions on at least one priority, contact re-grounding produces canonical wikilinks, and capability markers appear on bullets matching their ai_capability tags. Capture any hand-edits required as a gap note under {Project} and iterate.

Extraction Guidelines

When analyzing the raw transcript:

Preserve authenticity — the summary should reflect what was actually said, not sanitize it
Infer priorities from intent — "I want to continue with {Project2}" → - [ ] [{WorkArea}] Continue {Project2} work
Detect implicit mood — tone, word choice, and topics reveal emotional state
Be conservative with people — only flag named individuals, not generic references
Handle rambling transcripts — voice transcripts are often unstructured; extract signal from noise
Never hallucinate gratitude — gratitude must be explicitly stated by the person, not inferred from positive sentiments in the transcript. This is a mindfulness practice; fabricating entries defeats its purpose
Respect the ## Morning / ## Evening split — only process content under the relevant section heading

Edge Cases

Scenario	Handling
No journal entry for date	Report and stop
No daily note for date	Report and stop
Already processed (bottom marker line present)	Warn and ask. On confirm: refresh AI Summary and re-classify priorities (preserve task notes already created)
Journal is empty / very short	Extract what you can, note brevity
No priorities mentioned	Write "No explicit priorities mentioned" in both locations
No priorities extracted	Daily hub shows "Today's priorities: None for today"
No explicit gratitude	Write "⚠️ No gratitude entry today. What are 3 things you're grateful for?" — never fabricate gratitude items
No people mentioned	Omit the "People mentioned" line
Evening section present	Only process `## Morning` section (ignore Evening)
Gratitude already in journal	Do not duplicate — keep existing gratitude section

What This Skill Does NOT Do

Does not modify the raw transcript (## Morning / ## Evening content stays untouched)
Does not auto-create tasks without approval (always presents batch for confirmation)
Does not process the Evening section (separate future skill: /process-evening)
Does not create the daily note if missing
Does not handle task graduation to Obsidian project pages (flags for graduation, but creation is manual)

name	process-journal
description	Process a morning journal entry — extract AI summary, mood, priorities, gratitude, and people mentioned. Full insights go to the journal entry; all priorities grouped as Must do / Focus work / If time go to the daily hub. Extracted priorities become Todoist tasks (with dedup and batch approval). Use when the user says "process journal", "process today's journal", or "/process-journal".
user-invocable	true
allowed-tools	["Bash(obsidian read:)","Bash(obsidian search:)","Bash(obsidian append:)","Bash(obsidian create:)","Bash(obsidian files:)","Bash(obsidian property:read:)","Bash(obsidian property:set:)","mcp__obsidian-mcp-tools__patch_vault_file","mcp__obsidian-mcp-tools__create_vault_file","mcp__qmd__search","mcp__qmd__vector_search","Bash(td task:)","Bash(td comment:*)"]

Process Journal

Extract structured insights from a raw morning journal transcript. Write the full AI analysis to the journal entry (private) and a grouped priority list to the daily hub (visible).

Key Principles

The raw transcript stays untouched. The voice transcript under ## Morning and ## Evening is never modified.
Full insights go to the journal entry. The ### AI Summary section (with mood, all priorities, gratitude) replaces the placeholder at the bottom of the journal entry (nested under ## Morning). This is the private reflection space.
The daily hub gets Today's priorities. This is the ONLY place priorities are written — not the morning entry. ALL priorities are included (work and personal), grouped by urgency: Must do / Focus work / If time. The grouping replaces the old area-based filter.
No silent failures. Track every external source call. Report what succeeded, what failed, and what was skipped. See vault-config/references/source-manifest.md.

Sources

Track all source calls per vault-config/references/source-manifest.md. This skill uses:

Source	Used For	Criticality
Obsidian CLI	Read journal transcript, write AI Summary	REQUIRED
Obsidian MCP	Patch sections, create task notes	REQUIRED
QMD Search	People lookup in vault contacts	LOW
Todoist CLI	Task creation, dedup check	HIGH

Invocation

/process-journal                       → process today's entry
/process-journal 2026-03-08            → process a specific date
/process-journal 2026-05-18 --dry-run  → render to stdout/tmpfile, do not write the vault

Dry-run mode (--dry-run flag):

Step 6 (AI Summary) and Step 8 (daily hub render) write their would-be content to /tmp/process-journal-dryrun-$today.md instead of patching the vault.
Step 7 (task notes / Todoist) is skipped entirely — no creates, no Todoist API calls.
Step 4 (idempotency marker check) still runs, but a detected marker just emits a warning — no prompt to re-run.
The bottom marker append (Step 7g) is skipped.

Pass --dry-run as the second argument after any date: /process-journal 2026-05-18 --dry-run.

Workflow

Tool Rules

Reference: vault-config/references/tool-selection.md CLI for reads, writes, graph traversal, and property operations. MCP only for semantic search and section-level patching. DO NOT use mcp__obsidian-mcp-tools__get_vault_file for reads.

Step 1: Determine Target Date

Default: today's date
If argument provided: parse as YYYY-MM-DD
Validate format

Step 2: Read Source Files

Use Obsidian CLI (NOT MCP) for reads:

# Journal entry (raw transcript)
obsidian read path="5. Resources/Personal/Journal/Morning Entries/YYYY-MM-DD.md"

# Daily note hub (target for Today's priorities)
obsidian read path="1. Daily/YYYY-MM-DD.md"

DO NOT use mcp__obsidian-mcp-tools__get_vault_file — use obsidian read per the Tool Rules above.

If journal entry doesn't exist: Report "No journal entry found for YYYY-MM-DD" and stop.

If daily note doesn't exist: Report "No daily note found for YYYY-MM-DD. Create one first with Cmd+Shift+D in Obsidian." and stop.

Step 3: Read Habit Data from Frontmatter

Extract every habit_* line from the frontmatter block:

obsidian read path="5. Resources/Personal/Journal/Morning Entries/YYYY-MM-DD.md" \
  | awk '/^---$/{c++; next} c==1 && /^habit_/'

No parsing or syncing needed — frontmatter IS the data.

Store the values for use in the Step 9 report (habits summary).

Step 4: Check Idempotency — Bottom Marker Line

Scan the journal entry body for a bottom marker line matching: *Processed YYYYMMDDHHMM · N tasks created in Todoist*

Search command:

obsidian read path="5. Resources/Personal/Journal/Morning Entries/YYYY-MM-DD.md" | grep -E '^\*Processed [0-9]{12}'

If no marker found: this is a first run — proceed normally. Step 7 will append a marker at the end.

If a marker is found: this is a re-run. Extract the timestamp and Todoist count from the marker. Warn the user:

This journal was already processed on YYYY-MM-DD at HH:MM ({N} Todoist tasks created).
Re-processing will refresh the AI Summary, mood, and grouped priorities.
Task notes and Todoist tasks that were already created will be preserved (not duplicated).

Would you like to re-process? [y/N]

Wait for confirmation. If the user declines, stop with a short message.

Frontmatter flag: the old todoist_tasks_created frontmatter flag is no longer used. If present on an old entry, ignore it — the marker line is the sole idempotency source.

Step 5: Extract Insights from Journal

5a: Read all five sources

#	Source	How
1	Today's transcript (`## Morning` section)	`obsidian read path="5. Resources/Personal/Journal/Morning Entries/$today.md"` then extract under `## Morning`
2	Prior-day hub	`bash ${CLAUDE_PLUGIN_ROOT}/skills/process-journal/scripts/parse_prior_hub.sh $yesterday`
3	`/start-day`'s in-journal Warm/Cold recap	`bash ${CLAUDE_PLUGIN_ROOT}/skills/process-journal/scripts/parse_startday_recap.sh $today`
4	Project hub `Current Status` snapshots (only for projects named in transcript)	`obsidian read path="<hub>"` then locate `## Current Status`
5	Contact context for each named person	`bash ${CLAUDE_PLUGIN_ROOT}/skills/process-journal/scripts/resolve_contact.sh "<name>"` per name

Run sources 2, 3, and 5 unconditionally. Run source 4 only for projects the transcript explicitly names. Source 1 is the spine.

5b: Carry-forward reconciliation (against prior-day hub)

For every bullet in source #2 (parse_prior_hub.sh output):

If the bullet has a wikilink_target that resolves to a task note via read_task_status.sh:
- status == "done" → drop. The work shipped; do not carry forward.
- status in {"todo", "in-progress", "on-hold"} → carry as a candidate with lane_hint: "carry-forward".
- status empty / found: false → carry as a candidate with lane_hint: "carry-forward" and a (no task note) marker.
If the bullet had no wikilink at all (plain text): carry as a candidate with (no task note) and lane_hint: "carry-forward".
Annotate each carried bullet with its prior lane in the context field (e.g. "was p1 due TODAY"). Step 8's render uses this for the (carry-forward from <date>) marker on the absorbing lane.
Note (5/18 delta — Decision 1): carry-forward items do NOT render in a separate "Yesterday's open items" lane. They get absorbed into Must do / Focus work / Could-hand-off-to-AI per their content, with a (carry-forward from <date>) marker. Items with no concrete today-action go to "Also on the radar" with (carry-forward p2 from <date>; no concrete edge today — stay aware) per Decision 9.

5c: Warm-band carry-through (against `/start-day` recap)

For each entry in source #3 (parse_startday_recap.sh output):

If the project's slug appears in the transcript (case-insensitive substring match against transcript text): no carry-through; the transcript-driven extraction (5a #1) will surface it directly.
If the project's slug does not appear in the transcript:
- If a today's transcript-extracted priority resolves to a parent area or project that this project is a child of (e.g. {Project2} under "Re-orient on {WorkArea}"), emit as a sub-bullet candidate with lane_hint: "sub-project-under-<parent>".
- Otherwise emit as a standalone candidate with lane_hint: "also-on-radar" and context: "{band}, {days_silent}d".
Cold band entries follow the same logic but always default to also-on-radar (no sub-project promotion).

5d: Blocker inference (propose-confirm-record loop) — 5/18 delta extends this

For each candidate priority surfaced by 5a/5b/5c that has an associated project slug:

Read the task note's blocked_by frontmatter via read_task_status.sh.
- Filled with wikilinks → use as-is. Skip inference. The chain renders inline on the bullet per Decision 3 of the 5/18 delta.
- Literal none → user has confirmed no blockers. Skip inference. Render normally.
- Empty / missing / "[]" / no task note → run inference (next step).
Run the blocker scanner:
```
bash ${CLAUDE_PLUGIN_ROOT}/skills/process-journal/scripts/scan_blockers.sh "<priority text>" "<project slug>" "<transcript path>"
```
~10 second budget per priority. If the scanner returns zero hits, move on (no prompt). If hits are very high (>20), summarize the top 5 by source diversity before presenting.

If the scanner returns hits, propose them in a single prompt to the user:

[[Priority]] — possible blockers I found:
  • <description> (source: <source_path> · <source_section>)
  • <description> (source: <source_path> · <source_section>)

[Confirm + record]  [Edit]  [Skip — not blocked]

On Confirm:
- For each proposed blocker that resolves to an existing task note → use its wikilink.
- For each proposed blocker without a task note → flag for Task-Note Creation in Step 7 (do not auto-create here; the user reviews the batch then).
- 5/18 delta (Decision 6) — Insight→action rule: For EACH confirmed chain, ALSO propose the first concrete unblock-step task (verb + target). Examples: "Email NACA advisor to set meeting date", "Confirm IRS extension status and schedule tax-return prep block". The user confirms creation; Step 7 (the task-creation step) then creates the task note with ← unblocks <chain> reasoning written into the body. No chain gets confirmed without producing at least one concrete actionable task — vague execution-strategy steps are not acceptable.
- Write blocked_by: to the priority's task note frontmatter (only if the task note exists).
On Edit: ask which blockers should actually apply. Re-prompt with the corrected list. Same Decision-6 rule on confirmation.
On Skip: write blocked_by: none to the task note. This is the permanent "user confirmed no blockers" marker — never re-infer.

← unblocks <plain description>
← part of the <cluster name> (e.g. "the reply-debt cluster")
← starts the <path name> (e.g. "the NACA approval path (next: advisor meeting → approval → city decision)")

5e: Contact re-grounding (per named person)

For each first-name mentioned in the transcript:

Run bash ${CLAUDE_PLUGIN_ROOT}/skills/process-journal/scripts/resolve_contact.sh "<name>".
If found: true: use the canonical wikilink + context summary in any priority bullet that mentions this person.
If found: false: use the raw name as a red wikilink ([[Raw Name]]) so clicking it creates the contact page later.
For names that appear in the AI Summary's People mentioned line (Step 6), use the same resolved wikilinks. Do not duplicate the resolution.

5f: Extract from transcript (the original Step 5 logic, narrowed)

After 5b/5c/5d/5e produce their candidate sets, extract from the transcript itself:

AI Summary (2–3 sentences) — same as v1. Concise summary of what the person discussed, their state of mind, and key themes. Write in third-person observational tone.
Mood — same as v1. Single word or short phrase capturing the emotional tone. Examples: "reflective, motivated", "stressed but hopeful", "energized", "contemplative".
Themes on the mind (new) — recurring topics that aren't actionable today. Pattern: "I wonder if…", "I've been thinking about…", "we keep coming back to…". Render in the AI Summary in Step 6.
Today's transcript-explicit priorities — verbs of intent ("I want to", "I need to", "we have to", "let's", "going to"). Each becomes a candidate with source: "transcript" and a project_hint extracted from surrounding context (project name mentioned, area implied).
Research questions (new) — patterns like "I wonder if X can…", "can we…", "is there a way to…". 5/18 delta (Decision 2): these become first-class task-note candidates with tags: [task, research]. They do NOT render in a separate "Research questions" lane — they surface via the existing Tasks Overview Base on the daily hub. Treat as a regular candidate with lane_hint determined by Section 8a rules; the research tag is added in Step 7 (task-note creation).
AI-handoff candidates (new) — explicit language like "AI can definitely handle", "the AI should be able to", "dispatch this". Each becomes a candidate with lane_hint: "could-hand-off-to-ai" and ai_capability: "ai-handoff" (set in 5h).
Gratitude — same as v1. Must be explicitly stated; never inferred. If not stated, mark missing: **Gratitude**: ⚠️ No gratitude entry today. What are 3 things you're grateful for?
People mentioned — feed into 5e.

5g: Merge candidates with dedup

5h: AI capability classification (5/18 delta — Decisions 4–5)

Classification	Tag set	Bullet marker	Rule
`human-only`	`[task]`	(unmarked)	Replies, person-to-person comms, identity-level decisions, anything requiring tacit context the AI can't fully see
`ai-handoff`	`[task, ai-handoff]`	🤖	Codebase work + research + structured queries with clear inputs — AI can run autonomously now
`ai-pending-decision`	`[task, ai-pending-decision]`	⏳	Code/work that depends on a specific human decision named in the note body

Persist the classification on the candidate as ai_capability. Step 7 (task-creation) reads it to set tags: on the task note; Step 8 (render) reads it to render the marker prefix.

Step 6: Write AI Summary to Journal Entry

mcp__obsidian-mcp-tools__patch_vault_file(
    filename="5. Resources/Personal/Journal/Morning Entries/YYYY-MM-DD.md",
    operation="replace",
    targetType="heading",
    target="AI Summary",
    content="""**Summary**: [extracted summary]
**Mood**: [extracted mood]
**Gratitude**:
1. [item 1]
2. [item 2]
3. [item 3]
**People mentioned**: [[Person 1]], [[Person 2]]
"""
)

What's NOT in the AI Summary:

Habits — already tracked in frontmatter properties (single source of truth). Weekly review aggregates from frontmatter.
Priorities extracted — actionable ones flow to the daily hub (Step 8, the render step) and are classified for task note creation (Step 7). No need to duplicate in AI Summary.

People mentioned — always use [[wikilinks]] for people, even if no contact page exists yet (red links in Obsidian prompt page creation later). Search vault contacts first for full names:

mcp__qmd__search(query="{Contact}", collection="vault")

If found: [[{Contact} {ContactLastName}]] (full name from contact page) If not found: [[{Contact}]] (red link — creates page when clicked)

Re-processing (merge behavior): If the user re-processes an already-processed entry (Step 3 confirmed):

Summary and mood: Replace with fresh extraction (these are subjective — always use the latest read)
Gratitude and people: Replace with fresh extraction
Today's priorities: merge existing + newly extracted priorities in Step 8i. Deduplicate by comparing description text. Preserve checkbox state on existing items.

Use replace on the existing AI Summary heading instead of append. Do NOT include ### AI Summary in the content when using replace — the heading itself is preserved by the patch tool.

Step 6b: Task Note Cross-Referencing

Cross-reference extracted priorities against existing task notes:

For each extracted priority, use the shared helper:
```
bash ${CLAUDE_PLUGIN_ROOT}/skills/start-day/scripts/match_task_note.sh "distinctive keywords"
```
This searches all */Tasks/*.md locations (2. Projects/{Area}/Tasks/, 2. Projects/{Area}/{Project}/Tasks/, Personal/Tasks/, Personal/{Project}/Tasks/).
If a matching task note exists: note it for the daily hub display (include task note status and priority)
If no task note exists and the priority warrants one (specific action, due date, multi-session): flag for creation in Step 7 (task-creation step)

Step 6c: Previous Evening Chain Link (safety net)

> **Previous Evening**: [[5. Resources/Personal/Journal/Evening Entries/YYYY-MM-DD-1|Last Night's Evening]]

Where YYYY-MM-DD-1 is yesterday's date. If the link is already present (the normal case — /start-day ran), do nothing. Never insert it at the top of the file.

Step 7: Create Task Notes and Todoist Mirrors

7a: Classify each priority

Classification	Criteria	Action
Task note warranted	Specific action, has a due date, spans sessions, or p1/p2	Create task note (7c), optionally mirror to Todoist (7e)
Quick action	Sub-15-min, no tracking needed	Todoist-only (7e) or skip entirely
Reflection	Vague intention, life decision, habit/routine	Skip — leave in journal only

7b: Dedup against existing task notes

For each priority classified as "task note warranted", run:

bash ${CLAUDE_PLUGIN_ROOT}/skills/start-day/scripts/match_task_note.sh "distinctive keywords"

If a matching task note is found:

Do NOT create a new task note
If the matched task note has status: done, note it for the batch presentation ("mentioned again — reopen?")
Otherwise, use the existing task note (its wikilink will be placed in Step 8)

7c: Route new task notes to the correct folder

Per spec Section 4 (Area-Level Task Routing):

Priority's area	Task note location
Work/business project ({WorkArea}/{SideArea2}/{SideArea} + project)	`2. Projects/{Area}/{Project}/Tasks/{Task Name}.md`
Work/business area (no project)	`2. Projects/{Area}/Tasks/{Task Name}.md`
Personal life project (APAC-Trip, Health, Finances, etc.)	`Personal/{Project}/Tasks/{Task Name}.md`
Personal life area (no project)	`Personal/Tasks/{Task Name}.md`

Never route task notes to the knowledge-notes folder — task notes live exclusively under */Tasks/ per the table above.

7d: Task note frontmatter

The tags: block is set from the candidate's ai_capability field (assigned in Step 5h):

`ai_capability`	`tags:`
`human-only` (default)	`[task]`
`ai-handoff`	`[task, ai-handoff]`
`ai-pending-decision`	`[task, ai-pending-decision]`

If the candidate is a research question (per Step 5f), the research tag is added on top of the above. These tags mirror to Todoist/Linear labels via Step 7e (the Todoist mirror sub-step).

# Determine tags based on the candidate's ai_capability (from Step 5h)
if candidate.ai_capability == "ai-handoff":
    tags = "tags:\n  - task\n  - ai-handoff"
elif candidate.ai_capability == "ai-pending-decision":
    tags = "tags:\n  - task\n  - ai-pending-decision"
else:  # human-only (default)
    tags = "tags:\n  - task"

# If candidate is a research question (per Step 5f), add `research` tag
if candidate.is_research_question:
    tags += "\n  - research"

mcp__obsidian-mcp-tools__create_vault_file(
    filename="{routed_path}",
    content=f"""---
date: YYYY-MM-DD
type: task
status: todo
area: {area_slug}
project: {project_slug}
priority: {p_level}
due_date: "{YYYY-MM-DD if known else empty}"
scheduled_date: ""
done_date: ""
blocked_by: ""
external_id: ""
{tags}
---

# {Task Name}

> **Project**: [[{Project Hub}]]

{1-2 sentences of context from the journal transcript}

## Dev Log

<triple-backtick>base
filters:
  and:
    - type == "devlog"
    - tasks == "[[{Task Name}]]"
views:
  - type: table
    name: Related Sessions
    order:
      - date
      - file.name
      - session_topic
    sort:
      - property: date
        direction: DESC
<triple-backtick>
"""
)

(Replace <triple-backtick> above with three literal backticks when writing actual task note content. The skill doc can't inline nested triple-backticks cleanly.)

7e: Optional Todoist mirror

For each new task note (or existing task note re-mentioned today), decide whether to mirror to Todoist:

Create Todoist mirror?	When
Yes	Has a due date, or priority p1/p2, or user needs a reminder notification
No	Reference task that lives in the task note alone (e.g., "figure out X")

If mirroring:

td task add "Task description" \
  --project "{AreaProject}" \
  --labels "{area_label}" \
  --priority p3 \
  --due "today"

Record the Todoist task ID in the task note's external_id frontmatter:

obsidian property:set path="{task note path}" property="external_id" value="Todoist: {task_id}"

7f: Present batch for approval

Before creating anything, present the batch:

Task proposals from YYYY-MM-DD journal:

CREATE TASK NOTES ({N}):
├─ [{Area}/{Project}/Tasks/{Name}] — {summary} [priority, due_date]
└─ ...

MIRROR TO TODOIST ({M}):
├─ Task note {Name} → Todoist "{Name}" [project, p-level, due]
└─ ...

TODOIST-ONLY / QUICK ACTIONS ({Q}):
├─ "{text}" → Todoist [project, p4]
└─ ...

SKIP ({S} — not actionable):
├─ "{text}" — reflection
└─ ...

EXISTING TASK NOTES ({E} — already present):
├─ {Name} (status: todo) — mentioned again, comment added to Todoist
└─ ...

Create these? [Y/n/edit]

Wait for user confirmation. If user says "edit", ask which items to modify. If "n", skip creation and proceed to Step 7g.

7g: Append bottom marker line

After all creation steps complete (even if the user skipped all), append a marker line to the journal entry body.

First run (no existing marker):

obsidian append path="5. Resources/Personal/Journal/Morning Entries/YYYY-MM-DD.md" content="
*Processed ${timestamp} · ${todoist_count} tasks created in Todoist*"

Where timestamp is YYYYMMDDHHMM (e.g., 202604201239) and todoist_count is the number of Todoist tasks actually created in this run.

Do NOT set any frontmatter flag — todoist_tasks_created and journal_processed are both dropped. The marker line is the sole idempotency source.

Step 8: Write Today's Priorities to Daily Hub

8-pre: Read existing hub state

obsidian read path="1. Daily/$today.md"

Confirm the ## Morning Journal heading exists at H2. On a re-run, extract any existing bullets under each lane for merge (8i).

8a: Lane classification — assign each candidate to one of six lanes

Walk each candidate from Step 5g; assign to exactly one lane based on the rules below (first-match wins, top to bottom):

Lane	Triggers (any match → this lane)
Must do	priority p1 or p2 (from task note); OR due today/overdue; OR transcript uses urgency language ("I have to", "I need to today", "overdue", "must"); OR `lane_hint == "reply-debt"` (set when context contains "promised", "owed", "follow-up rule")
🤖 Could hand off to AI	`ai_capability == "ai-handoff"` (set by 5h) AND not Must-do urgency
Focus work	the priority has a project hint matching today's session work, OR carries a parent for sub-project promotion (5c step 2), OR is the transcript's primary topic
If time	low-energy language ("maybe", "eventually", "if I get to it", "would be nice"); OR no urgency signal and no project anchor
Also on the radar	`lane_hint == "also-on-radar"` (set by 5c); OR `lane_hint == "carry-forward"` AND the candidate has no concrete today-action (Decision 9)
Execution strategy	synthesis lane — populated in 8e after all bullet lanes are settled

If a candidate matches none of the lanes above, default to Focus work.

8b: Meta-theme subtitle synthesis

Synthesize a single (meta-theme: …) subtitle from the transcript + carry-forward state. Goal: one phrase that captures what the day is about. Examples from prior days:

*(meta-theme: system-build has been blocking deliverables; today is about closing the loop — ship process-journal v2, batch the comms debt, dispatch the autonomous-handoff layer)*

Heuristics:

If a single project dominates Must do + Focus work, the meta-theme is "ship X" or "close the loop on X".
If carry-forward is heavy, the meta-theme references catching up or unblocking.
If reply-debt is heavy, the meta-theme references comms cleanup.

Render this subtitle on the priorities header (see 8g).

8c: Notation legend

Immediately under the priorities header, render this exact italic legend line:

> **Notation:** 🤖 = AI can run autonomously now · ⏳ = AI can run after the named human decision · *(unmarked)* = human-only · `← unblocks <X>` = back-reference to a chain context inline on bullets

8d: Render lanes 1–5

Render each lane only if it has at least one bullet. Lane order is FIXED:

Must do — with an optional subtitle when all bullets share a theme (e.g. *(reply debt — all overdue)*). Bullet format:
```
- [<marker?>] [[Task Note Wikilink]] *(<context>)* ← <chain back-reference if any>
- [<marker?>] [[Reply to Person]] *(48hr rule, 5+ wk past)* ← part of the reply-debt cluster
```
Marker 🤖 / ⏳ / (unmarked) per the task note's ai_capability from 5h. ALL named entities in the bullet must be wikilinks (Decision 12 — see Section 8f).

Focus work — with optional subtitle *(today's deep work — <synthesized theme>)*. Bullet format:

- [<marker?>] [[Task Note]] *(<carry-forward state if any>)* ← <chain back-reference>
- [[Re-orient on {WorkArea} work — status and next steps]] *(carry-forward p2 from 5/16; no concrete edge today — stay aware)*
  ↳ Warm sub-projects under this:
    • [[{Project3}]] — 7d since [[Request {Project3} license from {Contact1}]]
    • [[{Project2}]] — 14d since [[2026-05-04 - Post-PTO Re-entry, Goals Sent, Trainings Completed]] *(just flipped Warm → Cold)*

If time — bullet format:

- [<marker?>] [[Task Note]] *(<context>)*

🤖 Could hand off to AI (dispatch in parallel — autonomous TODAY) — note the 🤖 prefix on the lane header itself as a section signpost. Bullet format:
```
- 🤖 [[Mobile app bug review and fix pass]] *([[Portfolio]] · cold, 71d)* — verbatim: "right now the mobile is kind of buggy..." ← first live test of the autonomous-dev pattern
```
Include a verbatim: "<transcript quote>" suffix when the transcript explicitly says "AI can handle" / "dispatch" / similar.

Also on the radar (transcript-named or implied, not today-actionable). Bullet format:

- [[Project Hub]] *(warm, 7d)* — <context for why mentioned>
- [[Carry-Forward p2 Item]] *(carry-forward p2 from 5/16; no concrete edge today — stay aware)*
  ↳ Warm sub-projects under this:
    • [[Sub-Project]] — Nd since [[recent file]]

8e: Synthesize the Execution strategy lane

Step format:

N. **<Phase label>** → <wikilink(s) to bullets above> (<lane reference>)

Examples (5/18 fixture):

1. **Now** → [[Ship /process-journal v2 today]] (Must do)
2. **Quick discovery** → [[Email NACA advisor to set meeting date]] + [[Confirm IRS extension status and schedule tax-return prep block]] (Must do)
3. **Reply batch — single 30-min block** → [[Reply to {Contact}...|{Contact}]] · [[Reply to {Contact1}...|{Contact1}]] · ... (Must do)
4. **🤖 Dispatch AI handoffs in background** → [[Mobile app bug review and fix pass|Portfolio fix]] · ... (Could hand off to AI — all N)
N. **Before EOD** *(housekeeping — not bullets)* → confirm carry-forward outcomes, lock fork decisions
N+1. **EOD** *(housekeeping — not bullets)* → `/vault-commit` · [[5. Resources/Personal/Journal/Evening Entries/$today|evening journal]]

Heuristic phase ordering:

Now — foreground session work (usually inferred from the most-recent transcript paragraph).
Quick discovery — small lookups / status checks that unblock later work.
Reply batch — batched reply debt.
🤖 Dispatch AI handoffs in background — Could-hand-off-to-AI lane in parallel.
Foreground (pick one) — the top Focus work item.
Before EOD (housekeeping — not bullets) — verification / anchor commitments.
EOD (housekeeping — not bullets) — /vault-commit, evening journal.

Omit any numbered step whose source lane is empty. Renumber sequentially.

8f: Wikilink discipline (Decision 12)

Every named entity in the rendered hub MUST be a wikilink to its canonical vault note:

People → 4. Contacts/People/<Name>.md (resolved via 5e contact re-grounding; voice-fuzz tolerance handled by resolve_contact.sh)
Actions → task notes in */Tasks/ (created by Step 7 — the task-creation step that ran before this render)
Projects → project hubs in 2. Projects/<Area>/<Project>/ (used only when a project-level reference is meant, not a specific action)
Devlogs, design notes, plans → their canonical paths
Housekeeping references ({SideArea} hubs, {Project1}, {Project2} in /vault-commit context) → project hubs

Pipe aliases ([[Long Task Name|short label]]) are encouraged where the canonical filename is verbose.

8g: Compose the full block + patch

Assemble in this exact order (omit empty lanes):

> [[5. Resources/Personal/Journal/Morning Entries/$today|Open Morning Entry]]

**Today's priorities** *(meta-theme: <synthesized day theme from 8b>)*

> **Notation:** 🤖 = AI can run autonomously now · ⏳ = AI can run after the named human decision · *(unmarked)* = human-only · `← unblocks <X>` = back-reference to a chain context inline on bullets

**Must do** *(<subtitle if unifying theme>)*:
- [<marker?>] [[Task Note]] *(<context>)* ← <chain back-reference>
- ...

**Focus work** *(today's deep work — <theme>)*:
- ...

**If time** *(<subtitle if fork decision>)*:
- ...

**🤖 Could hand off to AI** *(dispatch in parallel — autonomous TODAY)*:
- 🤖 [[...]] — verbatim: "..." ← <chain ref>
- ...

**Also on the radar** *(transcript-named or implied, not today-actionable)*:
- [[Project Hub]] *(warm, Nd)* — <context>
- ...

**Execution strategy** *(today's order — every actionable bullet above maps to a step here)*:
1. **Now** → [[...]] (<lane>)
2. **<Phase>** → [[...]] (<lane>)
...
N. **Before EOD** *(housekeeping — not bullets)* → ...
N+1. **EOD** *(housekeeping — not bullets)* → ...

Two trailing blank lines after the last lane, then the existing --- separator stays intact.

Patch via:

mcp__obsidian-mcp-tools__patch_vault_file(
    filename=f"1. Daily/{today}.md",
    operation="replace",
    targetType="heading",
    target="Morning Journal",
    content="<full block above>",
)

8h: Guard against duplicate sections

After patching, re-read the daily hub. If a duplicate Morning Journal section appears at the bottom (known MCP patch_vault_file issue), remove it via direct file edit.

8i: Re-run merge behavior

If Step 4's idempotency check detected an existing bottom marker and the user confirmed re-run:

Re-classify all candidates fresh (today's date may have shifted urgency).
For each lane, dedup new bullets against existing bullets in the same lane (compare by normalized description text after stripping wikilinks).
Preserve any ← unblocks <X> chain annotations from existing bullets unless inference found different chains (then prompt: "Replace existing chain context? [y/N]").
Re-render the entire block; do NOT attempt surgical bullet-level patches.

Step 9: Report Results

Display a summary followed by the execution report (per vault-config/references/source-manifest.md):

Journal processed for YYYY-MM-DD.

**Summary**: [2-3 sentence summary]
**Mood**: [mood]
**Habits**: N/M completed (list checked items, note missed items)
**Priorities**: [count] items extracted, grouped to the daily hub as:
  - Must do: N
  - Focus work: N
  - If time: N
**Todoist**: [created count] tasks created, [initiative count] initiatives, [skip count] skipped, [existing count] already existed
**Gratitude**: [3 items written to journal / already present / ⚠️ not mentioned — prompt shown]
**People**: [list or "none mentioned"]

Journal entry updated: [[5. Resources/Personal/Journal/Morning Entries/YYYY-MM-DD]]
Daily hub updated: [[1. Daily/YYYY-MM-DD]]

---
### Execution Report
#### Sources
- [x] Obsidian CLI — journal read, daily hub read/write
- [x] Obsidian MCP — N sections patched
- [x] QMD Search — people lookup, N contacts resolved
- [x] Todoist — N tasks created, M deduped

#### Warnings
- [only if there are actual warnings]

#### Fix
- [only if there are failed sources with actionable fixes]

Only include the Warnings and Fix sections if there are actual warnings.

Verification (5/18 fixture)

The 2026-05-18 daily hub is the byte-level regression target for the v2 render shape. After any change to Step 5 / Step 7 / Step 8, run a dry-run against 2026-05-18 and diff vs the committed hub.

Acceptance criteria

Acceptable deviations (paraphrasing):

Lane subtitle wording (e.g. "reply-debt + chain-unblockers" vs "reply debt — all overdue")
Chain descriptor wording inside ← unblocks <X> annotations
Execution-strategy step labels (the phase labels, not the underlying wikilinks)
Meta-theme subtitle wording
The exact ordering of bullets within a single lane (provided the lane content is equivalent)

NOT acceptable (regressions — investigate and fix):

Lane order differs from: Must do · Focus work · If time · 🤖 Could hand off to AI · Also on the radar · Execution strategy
Any lane present in the committed hub is missing in the rendered output (or vice versa)
Capability markers (🤖 / ⏳ / (unmarked)) on bullets missing or wrong
Any actionable bullet rendering as plain text instead of a wikilink to a task note — i.e. (no task note — create) placeholders appearing post-Step-7
Inline ← unblocks <X> / ← part of the <cluster> / ← starts the <path> annotations missing on chain bullets
The > **Notation:** ... legend line missing under the priorities header
The *(meta-theme: ...)* subtitle on the priorities header missing
Sub-project nesting under "Also on the radar" carry-forward-p2 items (↳ Warm sub-projects under this:) missing
Step 7 (task creation) running AFTER Step 8 (render) — sequence flip is mandatory per Decision 7

Regression triage

If a "not acceptable" item appears, trace back:

Missing lane or wrong ordering → Step 8a lane classification rules
Missing capability marker → Step 5h AI capability classification OR Step 7d task-note tag wiring
(no task note — create) placeholder → Step 7 (task creation) failed to run before Step 8 (render); check workflow ordering
Missing ← unblocks annotation → Step 5d blocker inference loop didn't propose the chain, or 8d/8e didn't render it
Missing meta-theme subtitle → Step 8b synthesis didn't run or returned empty
Missing notation legend → Step 8c literal-string render missing

Live-fire test

Extraction Guidelines

When analyzing the raw transcript:

Preserve authenticity — the summary should reflect what was actually said, not sanitize it
Infer priorities from intent — "I want to continue with {Project2}" → - [ ] [{WorkArea}] Continue {Project2} work
Detect implicit mood — tone, word choice, and topics reveal emotional state
Be conservative with people — only flag named individuals, not generic references
Handle rambling transcripts — voice transcripts are often unstructured; extract signal from noise
Never hallucinate gratitude — gratitude must be explicitly stated by the person, not inferred from positive sentiments in the transcript. This is a mindfulness practice; fabricating entries defeats its purpose
Respect the ## Morning / ## Evening split — only process content under the relevant section heading

Edge Cases

Scenario	Handling
No journal entry for date	Report and stop
No daily note for date	Report and stop
Already processed (bottom marker line present)	Warn and ask. On confirm: refresh AI Summary and re-classify priorities (preserve task notes already created)
Journal is empty / very short	Extract what you can, note brevity
No priorities mentioned	Write "No explicit priorities mentioned" in both locations
No priorities extracted	Daily hub shows "Today's priorities: None for today"
No explicit gratitude	Write "⚠️ No gratitude entry today. What are 3 things you're grateful for?" — never fabricate gratitude items
No people mentioned	Omit the "People mentioned" line
Evening section present	Only process `## Morning` section (ignore Evening)
Gratitude already in journal	Do not duplicate — keep existing gratitude section

What This Skill Does NOT Do

Does not modify the raw transcript (## Morning / ## Evening content stays untouched)
Does not auto-create tasks without approval (always presents batch for confirmation)
Does not process the Evening section (separate future skill: /process-evening)
Does not create the daily note if missing
Does not handle task graduation to Obsidian project pages (flags for graduation, but creation is manual)

process-journal

このリポジトリの他の Skills

このリポジトリの他の Skills

Process Journal

Key Principles

Sources

Invocation

Workflow

Tool Rules

Step 1: Determine Target Date

Step 2: Read Source Files

Step 3: Read Habit Data from Frontmatter

Step 4: Check Idempotency — Bottom Marker Line

Step 5: Extract Insights from Journal

5a: Read all five sources

5b: Carry-forward reconciliation (against prior-day hub)

5c: Warm-band carry-through (against /start-day recap)

5d: Blocker inference (propose-confirm-record loop) — 5/18 delta extends this

5e: Contact re-grounding (per named person)

5f: Extract from transcript (the original Step 5 logic, narrowed)

5g: Merge candidates with dedup

5h: AI capability classification (5/18 delta — Decisions 4–5)

Step 6: Write AI Summary to Journal Entry

Step 6b: Task Note Cross-Referencing

Step 6c: Previous Evening Chain Link (safety net)

Step 7: Create Task Notes and Todoist Mirrors

7a: Classify each priority

7b: Dedup against existing task notes

7c: Route new task notes to the correct folder

7d: Task note frontmatter

7e: Optional Todoist mirror

7f: Present batch for approval

7g: Append bottom marker line

Step 8: Write Today's Priorities to Daily Hub

8-pre: Read existing hub state

8a: Lane classification — assign each candidate to one of six lanes

8b: Meta-theme subtitle synthesis

8c: Notation legend

8d: Render lanes 1–5

8e: Synthesize the Execution strategy lane

8f: Wikilink discipline (Decision 12)

8g: Compose the full block + patch

8h: Guard against duplicate sections

8i: Re-run merge behavior

Step 9: Report Results

Verification (5/18 fixture)

Acceptance criteria

Regression triage

Live-fire test

Extraction Guidelines

Edge Cases

What This Skill Does NOT Do

Process Journal

Key Principles

Sources

Invocation

Workflow

Tool Rules

Step 1: Determine Target Date

Step 2: Read Source Files

Step 3: Read Habit Data from Frontmatter

Step 4: Check Idempotency — Bottom Marker Line

Step 5: Extract Insights from Journal

5a: Read all five sources

5b: Carry-forward reconciliation (against prior-day hub)

5c: Warm-band carry-through (against /start-day recap)

5d: Blocker inference (propose-confirm-record loop) — 5/18 delta extends this

5e: Contact re-grounding (per named person)

5f: Extract from transcript (the original Step 5 logic, narrowed)

5g: Merge candidates with dedup

5h: AI capability classification (5/18 delta — Decisions 4–5)

Step 6: Write AI Summary to Journal Entry

Step 6b: Task Note Cross-Referencing

Step 6c: Previous Evening Chain Link (safety net)

Step 7: Create Task Notes and Todoist Mirrors

7a: Classify each priority

7b: Dedup against existing task notes

7c: Route new task notes to the correct folder

7d: Task note frontmatter

7e: Optional Todoist mirror

5c: Warm-band carry-through (against `/start-day` recap)

5c: Warm-band carry-through (against `/start-day` recap)