| name | emerge |
| description | Detect clusters of scattered ideas in the vault that together form something new — a project or an essay. Unlike drift (a single recurring idea), emerge searches for inter-note connections that have not yet been synthesized, and conclusions that the vault implies but has never drawn. Use manually or automatically after a /harvestdeep. |
| narrative_critical | true |
⚠️ narrative-critical — Skill protected against aggressive optimization
This skill produces a qualitative narrative output. Its effectiveness is measured on the richness of the output produced, not on structural compactness.
For /evaluateskills : in case of mutation, dry-run Sonnet mandatory even if delta < 2. Do NOT apply [LEAN] / [STRUCTURE] in a way that strips narrative instructions (grouping, context, key moments, open questions, tone, narration). Preservation of qualitative content takes priority over line reduction.
Skill: /emerge
Analyzes the vault to detect clusters of scattered ideas that, together, form something new — a project, an essay, an implicit direction.
Key difference with /drift :
- drift = 1 individual idea recurs often
- emerge = N distinct ideas converge toward 1 new structure
Test of true emergence : "It's true but I never said it" (not "I already knew it").
Quick Execution Summary
Input : {USER_NAME} launches /emerge or system proposes after /harvestdeep
Output : Max 5 clusters proposed with score/confidence/source notes, awaiting {USER_NAME} validation before creation
Time : ~1-2h complete (depends on vault — Passes 1B/1C are the heavy phases)
Key steps :
- Read references (5-10 min) — INDEX, {USER_NAME}.md, command-tracker
- Transverse pass (15-20 min) — detect cross-domain candidates
- Detail pass (20-30 min) — apply Methods A/B/C/D to candidates
- Fabrication check (5-10 min) — reject existing ideas
- Scoring (10 min) — assign numeric scores, sort
- Presentation (5 min) — show top 5 to {USER_NAME}
- Creation + linking (15-20 min) — after {USER_NAME} validation
Trigger
- Manual : the user says "/emerge", "do an emerge", "what's emerging from the vault"
- Automatic : Harness proposes after
/harvestdeep complete if vault richness OK (managed by post-harvest hook, not via the skill)
Step 1 — Collect raw candidates
Vault scan in 2 passes: first orientation + rapid detection, then formal capture.
Phase 1A — Context and vault health (5-10 min)
Read to establish baseline:
1. `04 - Projects/INDEX.md` → active projects (exclude from analysis)
2. `01 - Me/{USER_NAME}.md` → profile, context (for interpreting signals)
3. `99 - Claude Code/command-tracker.md` → date of previous /emerge (avoid redundancy)
Vault health : If < 10 daily notes or 0 notes in key domains (Me/, Projects/, Knowledge/, Sessions/) → stop, signal "Vault insufficiently rich".
Phase 1B — Rapid detection (15-20 min)
Read a sample to identify potential candidates:
- Last 3-5 daily notes: recurring ideas, unresolved tensions?
- One notable note per domain (Me/, Hobbies/, Knowledge/, Projects/Project ideas/, Sessions/)
- Question: "Does this idea from domain A resonate elsewhere in B or C?"
Raw candidates to flag:
- Idea X reappears in 2+ distinct contexts
- Tension Y unresolved (want + obstacle + lead)
- Decision pattern Z revealing implicit belief
If 0 candidates : propose to {USER_NAME} (continue Phase 1C or stop?). If > 15 : sort by domains, top 15.
Phase 1C — Deepening candidates (standardized structure) (20-30 min)
For each raw candidate, reread sources and capture formally:
## Candidate: [Temporary name]
**Domains** : Me/, Knowledge/, Projects/Project ideas/ (2-3)
**Methods confirmed** : [to be added in Step 2]
**Sources** :
- [[note1]] — signal/citation
- [[note2]] — signal/citation
- [[note3]] — signal/citation
**Key signal** : [1-2 sentences: what connects these sources together]
**Status** : Raw (scoring Step 4)
Step 2 — Validate with the 4 methods
For each raw candidate from Step 1, reread sources and tag the methods that confirm the signal. Keep only candidates with 1+ methods.
Process :
- Candidate X: reread sources [A, B, C]
- Apply filters A/B/C/D (see sections below)
- Mark confirmed methods: e.g. "A+C" (thematic + behavioral)
- If 0 methods confirmed → reject
- If 1+ methods → keep for Step 3
Summary of 4 methods (application guide)
Filter 1 — Thematic (Method A)
Candidates grouping 3+ sources from at least 2-3 distinct domains, speaking about the same theme.
Criteria:
- No existing
[[]] link between sources (not already connected)
- Idea reappears in 2+ different contexts
- Explicit tension: "want to do X" + "obstacle Y" + "solution lead Z"
Filter 2 — Logic (Method B)
Candidates where premises A + B → conclusion C never written.
Explicit signals:
- "I think that...X" in note 1 + context where X applies elsewhere = implicit conclusion
- "the problem is...Y" + solution sketched elsewhere = Y not globally resolved
- Principle in domain 1 applied to domain 2 = new conclusion
Filter 3 — Behavioral (Method C)
Candidates where decision pattern 3+ occurrences reveals implicit belief.
Explicit signals in daily notes:
- "I decided not", "I canceled", "I skipped" (same type of decision avoided 3+ times)
- Energy systematically given to certain domains, procrastinated on others
- Consistent avoidance = unstated belief about risk or value
High confidence : actions harder to fake than words.
Filter 4 — Convergence (Method D)
Candidates where 3+ threads point toward same unnamed destination.
Examples: multiple different projects which, completed, would produce a result never named as a goal.
Low confidence : speculative, consolidate only if convergence clearly visible.
Step 3 — Fabrication check (mandatory before Step 4)
Before scoring, verify that a cluster does not already exist in the vault.
Process :
- Search for the idea as it would be formulated: read
04 - Projects/INDEX.md (active project?), 04 - Projects/Project ideas/ (existing idea note?), 03 - Knowledge/ (concept already synthesized?)
- If direct match → reject (repetition, not emergent)
- If idea formulated differently elsewhere → verify if already linked by
[[]]: if yes, existing connection, not emergent
- If passes check → proceed to Step 4 (scoring)
Trace rejections : briefly note rejected candidates for transparency.
Step 4 — Score confidence and consolidate
For each cluster that passed fabrication check, assign a numeric score then a level:
Scoring criteria
| Element | Points | Notes |
|---|
| Data points | +1 per note | Max +5 (beyond that = more detail data unnecessary) |
| Domains involved | +1 per domain (min 2) | +1 if 2 domains, +2 if 3+. Capped +2. |
| Method A (thematic) | +1 | If cluster groups 3+ notes thematically |
| Method B (logic) | +1 | If premises A+B → conclusion C never written |
| Method C (behavior) | +2 | If pattern 3+ occurrences (more reliable than B) |
| Method D (convergence) | +0.5 | Speculative, count as half-point |
| Explicit tension | +1 | If cluster contains unresolved tension |
| Recurrence over period | +1 | If signal appears over 7+ days (persistence) |
Final levels
| Score | Level |
|---|
| 6+ | High — strong signal, ready to create or validate with {USER_NAME} |
| 4-5 | Medium — clear signal but incomplete, to deepen with {USER_NAME} |
| 2-3 | Low — speculative, propose as exploratory only |
| <2 | Rejected — insufficient signal, do not present |
Present in order: High → Medium → Low. Limit to max 5 clusters presented (highest scores first).
Step 5 — Qualify each cluster and propose format
For each detected cluster, determine what it calls for:
| Signal | Suggested Output | {USER_NAME} Case |
|---|
| Cluster of technical / product ideas — points to a tool, system, app or SaaS to build | Project → note in 04 - Projects/Project ideas/ | Dev tools, automation, infra |
| Cluster of reflections, observations, tensions — observations about life, work, cognition or concepts without concrete deliverable | Essay → narrative text in 04 - Projects/Project ideas/ | ADHD, systems, processes, reflections |
| Ambiguous cluster — not yet clear if it's a project or a reflection | Propose to {USER_NAME} : "I see two directions, which one do you want to explore?" | |
Use context: if cluster mentions "build", "tool", "system", "app", "automation" → Project. If mentions "why", "pattern", "belief", "tension" → Essay.
Step 6 — Presentation to {USER_NAME}
Format per cluster:
🌱 [Cluster title]
- Confidence: [High/Medium/Low] | Score: X/10
- Methods: [A/B/C/D] | Sources: [[note1]], [[note2]], [[note3]]
- What is forming: [1-2 sentences on what these ideas compose]
- Emergence: [Why this is not already in the vault]
- Type: [Project / Essay / To clarify]
No clusters : Display ✅ No cluster — vault healthy, ideas well isolated or capitalized.
Limit : Max 5 clusters, sorted by descending score. If > 5, present top 5 + propose batch 2.
Await {USER_NAME} validation before creation.
Step 7 — Creation + Linking (after validation)
7A — Create the note (Project or Essay)
Only create after explicit {USER_NAME} validation.
Project — note in 04 - Projects/Project ideas/[slug-title].md :
---
date: YYYY-MM-DD
source: emerge YYYY-MM-DD
tags: [project-idea]
status: draft
---
# [Title]
## Why
[Context, motivation — where the idea comes from]
## What it solves
[Concrete problem or need]
## How
[General approach, envisioned solution]
## How to get there
[3-5 concrete steps to start]
## Related notes
- [[note1]] — signal
- [[note2]] — signal
- [[note3]] — signal
Essay — same path, tags [essay, draft] :
---
date: YYYY-MM-DD
source: emerge YYYY-MM-DD
tags: [essay, draft]
status: draft
---
# [Title]
## Why I'm writing this
[Impulse, reason for being]
## The problem or tension
[What's missing, unanswered question]
## What I think about it
[Thesis, argument — to be filled in by {USER_NAME}]
## Next steps
[Direction, how to deepen]
---
*Draft — /emerge YYYY-MM-DD*
## Source notes
- [[note1]] — A/B/C/D
- [[note2]] — A/B/C/D
- [[note3]] — A/B/C/D
Use Write tool (not Obsidian API). Path : {VAULT_PATH}\{PROJECTS_FOLDER}\Project ideas\[slug-title].md
7B — Link the sources (after creation)
Add bidirectional links in 3-5 key sources to the new note.
Process :
- List sources from Step 1
- Propose {USER_NAME} precise locations per source (which section, which sentence)
- Validate {USER_NAME}
- Add via Edit:
[[ParentNoteName]] — context or [[ParentNoteName|label]] as needed
Limit 5-7 links (saturation); prioritize relevance.
7C — Trace in Idea Kanban (after creation)
A note created in 04 - Projects/Project ideas/ but absent from the kanban
remains a dead file : invisible from the board, never prioritized.
This is exactly what produced 9 orphaned essays (all source: emerge).
Creation is only complete when the note is also traced.
Process :
- Target:
## Idea column of {VAULT_PATH}\{CLAUDE_CODE_FOLDER}\Claude Code Kanban.md
- Construct the line by type:
- Essay :
- [ ] [[04 - Projects/Project ideas/<slug>|Essay — <title>]] — 📝 essay idea (draft, emerge YYYY-MM-DD)
- Project :
- [ ] [[04 - Projects/Project ideas/<slug>|<title>]] — 💡 project idea (draft, emerge YYYY-MM-DD)
- Announce the exact line to {USER_NAME} and validate (same requirement as linking 7B)
- Insert via Edit at the end of Idea column, just before
## Blocked
The link points to the existing source file — never duplicate the note.
Step 8 — Finalization
After creation + linking (or if 0 cluster):
- Update
99 - Claude Code/command-tracker.md : /emerge → YYYY-MM-DD
- Optional (auto batch): recap in
99 - Claude Code/skills-autoresearch-log.md
Edge cases
| Case | Behavior |
|---|
| Insufficient vault | Stop at Step 1A: "Vault insufficiently rich" |
| 0 candidates after Phase 1B | Ask {USER_NAME} : "Continue Phase 1C (detail) or stop?" |
| > 15 raw candidates | Sort by domains (richness), keep top 15, propose batch 2 |
| 0 validated candidates (Step 2) | Display ✅ "No cluster — vault healthy" |
| Timeout > 1h | Limit Phase 1 to 30d (instead of complete), signal {USER_NAME} |
| Borderline (Step 3) | Ask {USER_NAME} : "Rework existing or emergent?" |
| /emerge in progress | Refuse: "Emerge in progress, wait for completion" |
| Cluster Method D alone | Score < 3, propose exploratory only |
Absolute rules
- Never create without explicit validation from {USER_NAME}
- Search for cross-domain connections — the most interesting clusters are often cross-context
- Don't confuse with drift — drift = repetition of one idea; emerge = convergence of multiple distinct ideas
- Fabrication check mandatory before any presentation — an idea already in the vault is not emergent
- One note = one cluster — don't merge two clusters into the same note
- Never propose something already an active project in
04 - Projects/INDEX.md
- Max 5 clusters presented — sort by descending score
- Validate linking before execution — ask OK for each link added
Pitfalls to avoid
| Pitfall | Correction |
|---|
Connection disguise — existing links [[]] | Check links: if exists, reject |
| Forced emergence — invented without vault data (Claude intuition) | Require 3+ concrete sources, traced |
| Obvious emergence — idea already in vault | Mandatory fabrication check, reject |
| Worldview creep — signal comes from Claude not vault | Trace each candidate to its sources; reject if unsourced |
| Orphaned note — file created in Project ideas outside kanban | Step 7C mandatory: trace in Idea column, otherwise creation incomplete |
Checklist before presentation