Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
Identify overlapping projects and merge them into consolidated notes
🔀 Merge Projects Skill
Detect overlapping or closely related projects and merge them into a single consolidated project note. Also critically evaluate all projects for novelty and feasibility, and flag or remove those that don't meet the bar for a publishable research contribution.
CRITICAL RULES
Never merge automatically — always present candidates and get user approval
Never delete source projects until merged project is verified
Never remove projects automatically — always explain why and get user approval first
Merged project inherits the highest status of its sources (seed < idea < active < writing < finished)
All paper links from source projects must be preserved in merged project
All MOC links from source projects must be preserved in merged project
Bidirectional links must be updated: papers that linked to old projects must link to merged project
Vault index must be updated after merge
WORKFLOW
1. Choose mode
Ask user:
Scan for candidates? (analyze all projects for overlap, then critically evaluate all projects)
Merge specific projects? (user names 2-3 projects to merge)
2a. Scan mode — Spawn analysis subagent
Use the Agent tool to spawn a subagent. Pass it the SCAN SUBAGENT PROMPT below verbatim.
The subagent reads all project files, scores overlap between pairs, and evaluates every project for novelty and feasibility. It returns a structured report. Wait for it to finish before proceeding.
Present the subagent's findings to the user and get approval before proceeding to merges or removals.
2b. Direct merge mode — Spawn read subagent
Use the Agent tool to spawn a subagent. Pass it the DIRECT MERGE SUBAGENT PROMPT below, with the project names filled in.
The subagent reads the specified project files in full and returns their content. Use the returned content to confirm overlap and proceed with the merge.
3. Execute merge (main session)
After user approves which projects to merge:
Read the full content of the source project files (if not already returned by subagent).
Create merged project by integrating content section by section:
Section
Merge strategy
Current Focus
Keep primary's; note secondary's angle if distinct
Seed Idea
Combine triggers from all sources
Current Working Idea
Synthesize into broader framing
Research Question
Merge into unified question; keep sub-questions if distinct
Hypothesis / Claim
Keep primary; add secondary as alternative hypothesis if different
Context & Motivation
Union of all motivations
Possible Approaches
Union of all approaches (deduplicate)
Research Ideas & Variants
Secondary project's main direction becomes a variant
Related Work
Union (deduplicate)
Relationship to Prior Work
Union (deduplicate)
Possible Experiments
Union (deduplicate)
Open Questions & Risks
Union (deduplicate)
Next Actions
Union; remove completed/redundant
Relevant Papers
Union of ALL paper links (no paper lost)
Decide on merged title:
If primary title is broad enough → keep it
If both titles are narrow → create new title capturing combined scope
Title must be concise and descriptive
Set merged status: Use highest status among source projects.
Write the merged project file.
4. Find and update all references — Spawn reference-finder subagent
After the merged project is written, spawn a subagent with the REFERENCE FINDER SUBAGENT PROMPT below (fill in the absorbed project names and merged project name).
The subagent greps across Reading/, MOCs/, and Projects/ to find every file that references the absorbed project(s). It returns a list of files and the specific link text to replace.
Use the returned list to make edits in the main session:
Replace [[Old Project Name]] with [[Merged Project Name]] in each listed file
5. Remove absorbed projects
Only after verification:
Confirm merged project file exists and is complete
Confirm all paper links updated
Confirm all MOC links updated
Confirm no remaining references to old project names (verified by subagent in step 4)
Then delete secondary (absorbed) project files — only with user approval.
6. Update vault index
Notify user to run /update_vault_index, or spawn update_vault_index subagent directly.
SCAN SUBAGENT PROMPT
Pass this prompt verbatim to the Agent tool:
You are a research project analyst for an Obsidian vault.
Vault root: {{VAULT_ROOT}}
STEP 1: Read vault index
Read .claude/vault_index.md to get all project names, statuses, and one-line descriptions.
STEP 2: Read all project files
For each project in Projects/:
- Read YAML frontmatter (status, type)
- Read "Seed Idea", "Current Working Idea", "Research Question", "Hypothesis / Claim", and "Relevant Papers" sections
- Do NOT read full file — only these sections
STEP 3: Identify merge candidates
For every pair of projects, assess overlap signals:
- Similar research questions or hypotheses
- Shared relevant papers (≥2 papers in common)
- Overlapping MOC links
- Complementary seed ideas (one project's approach is another's variant)
- Same phenomenon studied from slightly different angles
Score each pair:
- 🔴 High overlap: same core question, ≥3 shared papers, nearly identical hypotheses
- 🟡 Moderate overlap: related questions, 2 shared papers, one could be a variant of the other
- 🟢 Low overlap: distinct questions, ≤1 shared paper, different mechanisms (skip these)
Only report 🔴 and 🟡 pairs.
STEP 4: Critical evaluation — every project
For every project (not just merge candidates), assess:
Novelty:
- Does it propose something meaningfully different from existing published work?
- Is the contribution incremental to the point of being unpublishable at a top venue?
- Is the claimed insight already implicit or explicit in prior work?
Feasibility:
- Can the core hypothesis be tested with realistic resources?
- Is the method technically well-defined enough to implement?
- Are there fundamental blockers (proprietary model internals, unavailable datasets, unclear operationalization)?
Verdict per project:
- ✅ Keep — novel, feasible
- ⚠️ Borderline — minor concern; keep but flag
- ❌ Remove — fails novelty or feasibility; must state specific reason (2-4 sentences)
STEP 5: Return findings in EXACTLY this format:
MERGE_CANDIDATES:
🔴 [[Project A]] + [[Project B]]
Reason: <shared papers, similar question>
Suggested primary: [[Project A]]
Suggested merged title: <title if needed>
🟡 [[Project C]] + [[Project D]]
Reason: <...>
Suggested primary: [[Project C]]
Suggested merged title: <title if needed>
(or "none" if no candidates)
CRITICAL_EVALUATION:
✅ [[Project Name]]
✅ [[Project Name]]
⚠️ [[Project Name]] | <brief concern>
❌ [[Project Name]] | <specific reason why it fails novelty or feasibility bar>
(list every project)
DIRECT MERGE SUBAGENT PROMPT
Pass this prompt to the Agent tool (fill in project names):
You are a research project reader for an Obsidian vault.
Vault root: {{VAULT_ROOT}}
Projects to read: <PROJECT_NAME_1>, <PROJECT_NAME_2> (add more if needed)
For each project file in Projects/<name>.md:
- Read the FULL file content
Return the full content of each file verbatim, clearly separated by project name headers.
REFERENCE FINDER SUBAGENT PROMPT
Pass this prompt to the Agent tool (fill in absorbed project names and merged project name):
You are a reference finder for an Obsidian vault.
Vault root: {{VAULT_ROOT}}
Absorbed project(s) to find references to: <OLD_PROJECT_NAME_1>, <OLD_PROJECT_NAME_2>
Merged project name: <MERGED_PROJECT_NAME>
For each absorbed project name, search for all references across the vault:
Run (from vault root):
grep -rl "[[<OLD_PROJECT_NAME>]]" "Reading/" "MOCs/" "Projects/"
For each file found, extract the specific line(s) containing the old project link.
Return results in EXACTLY this format:
REFERENCES_TO_UPDATE:
File: Reading/<filename>.md
Line: [[Old Project Name]] → replace with [[Merged Project Name]]
File: MOCs/<filename>.md
Line: [[Old Project Name]] → replace with [[Merged Project Name]]
File: Projects/<filename>.md
Line: [[Old Project Name]] → replace with [[Merged Project Name]]
(or "NO_REFERENCES_FOUND" if nothing found)
TOKEN EFFICIENCY
Scan subagent reads all project files in isolation — does not pollute main context
Reference finder subagent uses grep, no file reads — fast and cheap
Direct merge subagent returns raw file content for main session to synthesize
Main session handles only: user interaction, content synthesis, file writes
OUTPUT RULES
Scan results: candidate pairs with overlap score + rationale, then full evaluation table
Confirm paper count preserved: "X papers linked before → X papers linked after"
SELF-CHECK
✅ Did I spawn the scan subagent before presenting findings?
✅ Did I get user approval before merging?
✅ Did I critically evaluate every project for novelty AND feasibility?
✅ Did I provide a brief explanation for every ❌ project before flagging it for removal?
✅ Did I get user approval before removing any project?
✅ Are ALL paper links from source projects preserved in merged project?
✅ Are ALL MOC links preserved?
✅ Did I spawn the reference finder subagent after the merge?
✅ Did I update all files returned by the reference finder?
✅ Is the merged project status correct (highest of sources)?
✅ Did I delete absorbed/removed project files only after full verification and user approval?
✅ Is the vault index updated or user notified to update?
✅ No orphan references remain to deleted project names?
