Cognos

A Multi-Platform Cognitive Plugin Suite for AI Agents

Cognos gives AI agents persistent memory, a knowledge graph, automated processing pipelines, skill generation from conversation patterns, and a self-evolving hivemind where every agent interaction makes the collective smarter. It runs across Claude Code, Claude.ai, Cursor, Codex, and Gemini CLI.

---

Quick Start (2 minutes)

# Clone the repo
git clone https://github.com/your-org/cognos.git
cd cognos

# Run the installer (auto-detects your platform)
bash platforms/universal/install.sh

# Start the onboarding
# In your AI tool, run:
/cognos:setup

What's Inside

Cognos is composed of 6 independent, composable modules that communicate through the filesystem (markdown files), not APIs.

Module	What It Does	Key Commands
cognos-memory	Persistent memory across sessions	/cognos:remember, /cognos:recall, /cognos:consolidate
cognos-graph	Knowledge graph with atomic notes & MOCs	/cognos:link, /cognos:map, /cognos:orphans
cognos-pipeline	6-phase processing pipeline (6R)	/cognos:inbox, /cognos:pipeline, /cognos:review
cognos-skillforge	Generate skills from conversation patterns	/cognos:forge, /cognos:forge:detect, /cognos:forge:test
cognos-identity	Agent identity & progressive onboarding	/cognos:setup, /cognos:identity, /cognos:context
cognos-hivemind	Self-evolving agent orchestration	/cognos:dispatch, /cognos:contribute, /cognos:evolve, /cognos:swarm

Install what you need. Each module works independently or as part of the full system.

Architecture

+-----------------------------------------------------+
|                  PLATFORM ADAPTERS                   |
|  Claude Code | Claude.ai | Cursor | Codex | Gemini  |
+-----------------------------------------------------+
|                   PLUGIN MODULES                     |
|  Memory | Graph | Pipeline | SkillForge | Identity  |
|  +--- Hivemind (orchestration layer) --------+      |
+-----------------------------------------------------+
|                    SHARED CORE                       |
|  Schema | Markdown Engine | Link Resolver | Config  |
+-----------------------------------------------------+

Layer 1: Shared Core

Pure markdown + YAML. No runtime dependencies. No database. The filesystem is the database.

Layer 2: Plugin Modules

5 standalone modules, each with its own SKILL.md, scripts, and agent prompts. Modules communicate through shared directories.

Layer 3: Platform Adapters

Each adapter translates Cognos conventions into platform-specific formats.

Directory Structure

cognos/
+-- cognos.yaml              # System configuration
+-- core/                    # Shared core (schema, engine, resolver, config)
+-- modules/
|   +-- cognos-memory/       # Persistent memory engine
|   +-- cognos-graph/        # Knowledge graph engine
|   +-- cognos-pipeline/     # 6R processing pipeline
|   +-- cognos-skillforge/   # Skill generator
|   +-- cognos-identity/     # Agent identity & context
+-- platforms/
|   +-- claude-code/         # Claude Code adapter + hooks
|   +-- claude-ai/           # Claude.ai project adapter
|   +-- cursor/              # Cursor adapter + .cursorrules
|   +-- codex/               # Codex adapter
|   +-- universal/           # Generic adapter + installer
+-- self/                    # Agent identity, memory, goals, facts
+-- notes/                   # Knowledge graph (atomic notes, MOCs)
+-- ops/                     # Sessions, inbox, logs, queue
+-- skills/                  # Generated skills live here

Platform Compatibility

Feature	Claude Code	Claude.ai	Cursor	Codex	Gemini
Memory hooks	Native	Via projects	Via rules	Via config	Via rules
Subagent spawning	Yes	No	Limited	Yes	No
File system access	Full	None (artifact)	Full	Full	Full
Git integration	Yes	No	Yes	Yes	No
Skill hot-reload	Yes	Manual	Yes	Yes	Manual

The 6R Pipeline

The pipeline transforms raw information into structured knowledge:

1. Record - Capture raw material into inbox
2. Reduce - Extract atomic insights
3. Relate - Find connections to existing notes
4. Reflect - Synthesize patterns across connected notes
5. Reweave - Update older notes with new context
6. Review - Quality check (schema, orphans, staleness)

Each phase runs as a subagent with fresh context to prevent attention degradation.

SkillForge: Skills That Create Skills

The flagship module. SkillForge monitors your conversations for repeatable patterns and converts them into installable, distributable skills.

Process:

1. Pattern Detection - Identifies repeatable workflows (2+ occurrences)
2. Skill Extraction - Converts pattern into SKILL.md with triggers, procedure, anti-patterns
3. Skill Testing - Evaluation loop (executor -> grader -> comparator -> analyzer)
4. Skill Refinement - Iterates until quality threshold (80+/100)
5. Skill Packaging - Cross-platform distribution-ready package

Hivemind: The Self-Evolving Brain

The Hivemind is Cognos's orchestration layer. It connects all modules into a collective intelligence that improves with every agent interaction.

How it works:

Agent receives task
    -> Hivemind dispatches best skill from the graph
        -> Agent executes using skill + context
            -> Agent contributes feedback (improvements, edge cases, ratings)
                -> Hivemind evolves the skill when consensus is reached
                    -> Next agent gets a better skill

Key capabilities:

• Dispatch — Match any task to the best available skill via graph search
• Contribute — Every agent submits improvements after using a skill
• Evolve — Skills auto-upgrade when 2+ agents confirm an improvement
• Swarm — Decompose complex tasks across parallel agents, each with their optimal skill

The collective effect: One agent discovers an edge case. The next agent gets a skill that already handles it. Every interaction raises the floor for the entire system.

Technical Decisions

Decision	Choice	Rationale
Storage	Plain markdown + YAML frontmatter	Max portability, no vendor lock-in
Database	None (filesystem)	ripgrep handles search
Config	Single cognos.yaml	One file to understand the system
Module communication	Filesystem (shared dirs)	No IPC, no APIs, no coupling
Skill format	Standard SKILL.md	Cross-platform compatible
Testing	Evaluation framework	Grading + comparison built in

Configuration

All configuration lives in cognos.yaml at the vault root:

version: "0.1.0"
platform: claude-code
modules:
  memory: true
  graph: true
  pipeline: true
  skillforge: true
  identity: true
spaces:
  self: "self/"
  notes: "notes/"
  ops: "ops/"
  skills: "skills/"

Contributing

Cognos is open source. Contributions welcome.

1. Fork the repo
2. Create a feature branch
3. Make changes (follow existing patterns)
4. Run the review pipeline: /cognos:review
5. Submit a PR

License

MIT

---

Built by Eli5DeFi | February 2026