By name: cognos-pipeline description: "The 6R Processing Pipeline — transforms raw information into structured, connected knowledge through six sequential phases." triggers: - /cognos:inbox - /cognos:reduce - /cognos:relate - /cognos:reflect - /cognos:reweave - /cognos:review - /cognos:pipeline platform: universal version: 0.1.0 depends: - cognos-memory - cognos-graph
Cognos Pipeline — The 6R Processing Pipeline
Purpose
Raw information is worthless until it is processed into knowledge. The 6R Pipeline is a disciplined, sequential workflow that takes any raw input — a URL, a conversation excerpt, a file, a stray thought — and transforms it into atomic, connected, retrievable knowledge inside the Cognos vault.
Each phase is designed to run as an independent subagent with fresh context. This prevents attention degradation: no single agent accumulates enough state to lose focus. The orchestrator hands off between phases, passing only structured artifacts (file paths, frontmatter) rather than full content.
The 6R Pipeline
| Phase | Name | Action | Input | Output |
|---|---|---|---|---|
| 1 | Record | Capture raw material into inbox | Content, URL, file, or stdin | ops/inbox/<item>.md |
| 2 | Reduce | Extract atomic insights | Inbox item path | notes/atomic/<note>.md (one per insight) |
| 3 | Relate | Find connections to existing notes | Atomic note path | Updated [[links]], MOC entries |
| 4 | Reflect | Synthesize patterns across connected notes | Domain or topic scope | Synthesis notes, updated MOCs |
| 5 | Reweave | Update older notes with new context | Newly processed note path | Backlink annotations on older notes |
| 6 | Review | Quality check the vault | Entire vault | Health report at ops/logs/review-<date>.md |
Commands
/cognos:inbox <content>
Quick capture. Drops content into the inbox with minimal friction. Alias for Record phase.
/cognos:reduce <inbox-item-path>
Extract atomic insights from a raw inbox item. Creates one note per distinct insight.
/cognos:relate <note-path>
Scan the vault for connections to the given note. Auto-links high-confidence matches. Updates MOCs.
/cognos:reflect <scope>
Synthesize patterns across a domain or topic. Scope can be a domain name, topic tag, or all.
/cognos:reweave <note-path>
Propagate new knowledge backward — update older notes that should reference the new material.
/cognos:review
Run a full vault health check. Produces a structured report with actionable recommendations.
/cognos:pipeline <source>
End-to-end processing. Runs all 6 phases sequentially on the given source material.
Subagent Architecture
The pipeline is built on a core principle: fresh context per phase.
Orchestrator
├── spawn → Record Agent (captures raw input, writes inbox item)
├── spawn → Reduce Agent (reads inbox item, writes atomic notes)
├── spawn → Relate Agent (reads atomic note, finds connections)
├── spawn → Reflect Agent (reads scope, synthesizes patterns)
├── spawn → Reweave Agent (reads note, updates older notes)
└── spawn → Review Agent (reads vault, generates health report)
Each agent receives only what it needs:
- Record: the raw content and a target path
- Reduce: the inbox item file path
- Relate: the atomic note file path and a list of existing note paths
- Reflect: a scope identifier and relevant note paths
- Reweave: the new note path and paths of potentially related older notes
- Review: the vault root path
No agent inherits the accumulated context of previous phases. This is intentional. Attention is finite. By resetting context between phases, each agent operates at peak focus on its specific task.
Queue System
The pipeline uses a queue file at modules/cognos-pipeline/queue/queue.md to track processing state. Items move through: Pending → In Progress → Completed (or Failed).
File Conventions
- Inbox items:
ops/inbox/YYYY-MM-DD-HHmm-<slug>.md - Atomic notes:
notes/atomic/<slug>.md - Synthesis notes:
notes/topics/<slug>.mdornotes/domains/<slug>.md - Pipeline logs:
ops/logs/pipeline.log - Review reports:
ops/logs/review-YYYY-MM-DD.md
Anti-Patterns
Do not skip phases
Each phase builds on the output of the previous one. Skipping Relate means your notes are isolated islands. Skipping Reflect means you never synthesize. Skipping Reweave means old knowledge never learns about new knowledge. Run the full pipeline.
Do not process too many items at once
The pipeline is designed for deliberate, focused processing. Batch sizes above 5 items risk shallow treatment. If you have 20 inbox items, process them in batches of 3-5, allowing Reflect to find patterns within each batch before moving on.
Do not let the inbox grow unbounded
An inbox with 50+ unprocessed items is a symptom of capture without processing. The inbox is a buffer, not a warehouse. Process regularly or stop capturing until you catch up.
Do not manually edit pipeline artifacts mid-process
If the pipeline is running, let it finish. Manual edits to inbox items or atomic notes during processing can cause the Relate and Reweave phases to operate on stale data.
Do not merge Reduce output
Each insight gets its own atomic note. If two insights feel related, that is what Relate is for. Premature merging destroys granularity and makes future synthesis harder.