consolidate

name	consolidate
description	Review inbox and recent notes, synthesize and merge into the knowledge base. Use when the user wants to process, organize, or clear the inbox regardless of topic — 'clean up my inbox', 'process my notes'. Do NOT use for: capturing new information now (use capture or braindump), synthesizing already-vaulted notes on a specific topic (use distill — consolidate works on the inbox), or finding note connections (use connect).
schedulable	false

Consolidate

Process your inbox and recent captures into your permanent knowledge base.

---

Step 1: Survey the Inbox

List all files in [inbox_folder]/ (excluding .gitkeep). For each file:

• Read the title and first few lines
• Note the date and main topics

Report: ────────────────────────────────────────────────────────────── 📥 Consolidate — {N} inbox items ──────────────────────────────────────────────────────────────

1. {filename} — {brief description}, {N tasks} tasks
2. {filename} — {brief description}

---

Step 2: Let User Choose Scope

AskUserQuestion:

• question: "Which items do you want to process?"
• header: "Consolidate Scope"
• multiSelect: false
• options:
  • label: "all", description: "Process all inbox items"
  • label: "last N days", description: "Process items from the last N days (specify N)"
  • label: "specific file", description: "Process a specific file (name it)"
  • label: "just review", description: "Review without moving anything"

---

Step 2.5: Pre-classify (parallel)

Before processing, dispatch one Inbox Classifier agent (agents/inbox-classifier.md) per selected inbox note in parallel (run_in_background: false, mode: "bypassPermissions"). Pass each note's note_path, note_content, vault_root, knowledge_folder, resources_folder, areas_folder, and projects_folder. Wait for all results before proceeding to Step 3.

Store each result as the default routing recommendation for that note. If a classifier call fails or returns an empty result, proceed without a recommendation for that note.

---

Step 3: Process Each Selected Item

For each item:

3a. Analyze

Read the file fully. Use the pre-classification from Step 2.5 as the starting point. Confirm or adjust based on your own reading:

• What type of knowledge this is (insight, reference, idea, project note, area)
• What existing notes it relates to (search via the search tools if available, otherwise Glob [knowledge_folder]/**/*.md, [resources_folder]/**/*.md, [projects_folder]/**/*.md, [areas_folder]/**/*.md). For search-tool results, only keep candidates with rerank_score ≥ 0.30 (prefer ≥ 0.60) — drop anything below 0.30.
• Whether it deserves its own note or should be merged into an existing one

3b. Decide Destination

Primary signal (check first): If the inbox item has a source: frontmatter field matching /research, /summarize, or /reading-notes, route it directly to [resources_folder]/ : no judgment needed.

Secondary signal (for all other notes): Apply the content-type rule below.

Classify the item and route it to the appropriate folder:

• Your own synthesis, insight, or conclusion → [knowledge_folder]/[subfolder]/
• Reference material, external info, or source notes → [resources_folder]/[subfolder]/
• Project-specific work → [projects_folder]/[subfolder]/
• Ongoing responsibility (something you maintain over time, not a one-time insight) → [areas_folder]/ : examples: health tracking, finances, career development, relationships

Confirm routing with the user for the first 3 items. After that, proceed autonomously : or if the user says 'stop and confirm', return to confirmation mode for the next item.

Why front-load confirmations: The first 3 items calibrate the model's understanding of the user's routing preferences. After that pattern is established, autonomous routing reduces friction. Front-loading prevents the worst mistakes (wrong folder for the first batch) without slowing down the whole session.

[{n}/{N}] {filename} looks like {classification} Route to {folder}/{subfolder}/?

Then AskUserQuestion:

• question: "How should I file this note?"
• header: "Route [{n}/{N}]"
• multiSelect: false
• options:
  • label: "confirm", description: "File as suggested"
  • label: "different folder", description: "Choose a different destination"
  • label: "merge", description: "Merge into an existing note instead"
  • label: "skip", description: "Leave in inbox for now"

Also show merge options if relevant:

I'd merge this into "Existing Note" : it adds context about [topic]. Or I could create a new note: "New Note Name". What do you prefer?

Mixed-content notes: If a single inbox item contains content that belongs in multiple folders (e.g., a braindump with both personal insights and project tasks), offer to split it: create separate notes for each content type, each routed to its correct folder. Ask the user to confirm before splitting.

3b.5 Choose Subfolder (for new notes only)

Based on the routing decision above:

1. Glob existing subfolders in the target folder ([knowledge_folder]/*/, [resources_folder]/*/, [projects_folder]/*/, or [areas_folder]/*/)
2. Suggest a kebab-case subfolder based on the note's topic (max 2 levels, e.g. science/biology)
3. Present to user: "I'd file this under [destination-folder]/[suggested-path]/. OK?"
4. Use confirmed path for file creation.

3c. Execute

• Merge: Append the content as a new section in the target note, with a date header. When merging, confirm the merge target is in the same folder as the routing decision. If the best merge target is in a different folder, note this to the user and ask which should take precedence: the routing decision or the merge.
• New note: Create [destination-folder]/[subfolder]/[Topic Name].md with proper frontmatter
• Archive: If the item is outdated or irrelevant, move to [archive_folder]/YYYY/MM/

Add wikilinks connecting to at least one related note found in 3a that cleared the rerank_score ≥ 0.30 threshold (Grep-fallback candidates are unaffected). If none cleared it, omit ## Related rather than linking a weak match.

---

Step 4: Handle Tasks

If inbox items contain unchecked tasks (- [ ]):

• Leave them in place (the Tasks plugin will find them wherever they live)
• Or ask: "Do you want to move the open tasks to your project notes?"

---

Step 5: Archive Processed Items

After an inbox item has been fully processed and its content merged/filed:

• Move the original file to [archive_folder]/YYYY/MM/ (using today's date, don't delete it)
• Or delete it if the user prefers a clean inbox

Ask preference once: "After processing, should I archive originals or delete them?"

---

Step 6: Summary

Report: ────────────────────────────────────────────────────────────── 📥 Inbox Processed — {N} notes ────────────────────────────────────────────────────────────── Moved: {filename} → {folder}/{subfolder}

Kept in inbox ({M}): {filename} — {reason} (omit "Kept in inbox" block if all items moved)

{P} tasks remain open across your vault. (omit if no open tasks) ────────────────────────────────────────────────────────────── {N} moved, {M} kept. Inbox {clear / down to M items}.

Empty state: ✅ Inbox is empty — nothing to process.

onebrain search reindex

---

Step 7: Write Log Entry

Follow ../_shared/audit-log-format.md (canonical frontmatter, append-per-day algorithm, run-section heading, failure mode) with:

• Filename: YYYY-MM-DD-consolidate.md — one file per day.
• Tags: [audit-log, consolidate]
• Skill: /consolidate
• Empty inbox: if Step 1 found nothing to process, skip writing — there is nothing to log.

Per-skill body template (canonical ## Run HH:MM heading; metadata in first bullet):

## Run HH:MM

- Inbox files processed: N

### Moved
- `[inbox_folder]/2026-05-10-ai-thoughts.md` → `[knowledge_folder]/ai/AI Architecture Thoughts.md` (new)
- `[inbox_folder]/2026-05-09-meeting-notes.md` → appended to `[projects_folder]/example/Project Name.md`

### Wikilinks added
- `03-knowledge/ai/AI Architecture Thoughts.md` ↔ `[[Agent Frameworks]]`, `[[OneBrain CLI]]`

### Skipped
- `00-inbox/2026-05-08-temp.md` — too short, asked user → kept in inbox

---

Progress reporting

This skill is long-running. Emit a 1-line status update after each major step so the user can see progress in real time.

In-session format:

→ [step N/M] <action being taken>

Examples:

→ [step 1/5] reading inbox · 8 notes found...
→ [step 2/5] dispatching inbox-classifier in parallel...
→ [step 3/5] classifier returned recommendations...
→ [step 4/5] applying user decisions per note...
→ [step 5/5] moving notes + updating MEMORY-INDEX...

Rules:

• Emit one line per major step (NOT per sub-step or tool call)
• M = total steps known up front (count them before starting)
• Status lines use → [step N/M] prefix exactly so they're visually distinct from skill output
• Do NOT emit heartbeats for fast operations (< 5 seconds)

---

Known Gotchas

• Mixed-content notes. Braindumps often start with a personal insight but contain project tasks, external references, and reflections all in one file. Read the FULL note before classifying — the first paragraph can be misleading about the overall content type.

• Tasks survive moves but can be duplicated by merges. When merging an inbox note into a destination, verify the write operation preserves unchecked - [ ] lines from the source note rather than dropping them. The Tasks plugin finds them wherever they live, but only if they aren't silently lost during the merge.