vault-pipeline

name	vault-pipeline
description	5-phase knowledge vault processing pipeline: seed → extract → reflect → reweave → verify. Uses Ralph Orchestrator for resumable queue-based execution.
version	1.0.0
owner_squad	mega-brain
megabrain_tier	Tier2
context	inline
agent	general-purpose
user-invocable	true
argument-hint	[source-path] [type: claim|enrichment|source]

Vault Pipeline — 5-Phase Knowledge Processing

Processes a knowledge source through 5 sequential phases, each in isolated context via Ralph Orchestrator. Fully resumable — interrupted sessions pick up from queue state.

Phases

Phase 1: Seed

• Detect source type (file, URL, session transcript)
• Create queue entry in .synapse/queue/queue.yaml via ralph.enqueue()
• Move source to .synapse/queue/archive/{batch}/

Phase 2: Extract

• Read source from archive
• Extract atomic notes (one claim per note)
• Each note follows insight-note.md template: description, scope, confidence, type, areas, relevant_insights
• Write to .synapse/vault/insights/ with prose titles (claim as title)

Phase 3: Reflect (Dual-Discovery)

• For each new note, run dual-discovery (see .synapse/methodology/dual-discovery.md):
  1. MOC navigation: read domain maps → find related notes
  2. Semantic search: grep for keywords in .synapse/vault/insights/
• Apply articulation test to each candidate connection
• Add bidirectional links for passing connections

Phase 4: Reweave

• Backward pass: for EXISTING notes, ask "if written today with everything we now know, what changes?"
• Actions: sharpen claim (vague → specific), split note (1 → N), challenge claim (new evidence contradicts)
• Record reweave trace

Phase 5: Verify

• Articulation test on all new connections
• Link health check (no dangling [[links]])
• MOC coherence (all new notes appear in at least 1 domain map)
• Discovery trace completeness (was dual-discovery executed?)

Usage

/vault-pipeline docs/sessions/2026-04/session-notes.md source
/vault-pipeline "my observation about capability routing" claim

Queue Integration

Uses .synapse/orchestration/ralph.js for phase tracking. After each phase, MUST output:

=== RALPH HANDOFF ===
Phase: {phase_name}
Work Done: {summary of what was created/modified}
Files Modified: [list]
Queue Update: advance to next phase
Learnings: {any insights about the process itself}
===================

Dependencies

• .synapse/orchestration/ralph.js (queue management)
• .synapse/vault/templates/insight-note.md (note template)
• .synapse/methodology/dual-discovery.md (dual-discovery methodology)
• .claude/skills/arscontexta-extract/SKILL.md (extraction methodology)