Codified Orchestrator (CCO) Documentation

CLI-first coding orchestrator for codified-context repositories. Full alignment with arxiv:2602.20478.

What is CCO?

Codified Orchestrator (CCO) is a CLI tool that routes coding tasks through specialized agents, retrieves only relevant context, and produces bounded local changes. It's built on the architecture described in arxiv:2602.20478 - "Codified Context: Infrastructure for AI Agents in a Complex Codebase".

Key Capabilities

19 Specialized Agents - From implementer to security-reviewer, each agent has focused prompts and trigger patterns
3-Tier Memory Architecture - Hot (AGENTS.md), Cold (.context/), and AOMA long-term storage
MCP Server - Model Context Protocol for agent communication
Task Lifecycle - Branch-based task isolation with start/finish/abandon workflow
Pre-flight Checks - Failure prevention before task execution

Project Status

v1.0.0

Current Version

552

Tests Passing

Specialized Agents

100%

Paper Alignment

Quick Start

ℹ️

Make sure you have Python 3.10+ and pip installed before starting.

1. Install CCO

pip install -e .
cco --version

2. Configure

cco config init --profile direct-minimax

3. Run Doctor

cco doctor --repo /path/to/your/repo

4. Execute a Task

# Plan first
cco plan --repo /path/to/repo --task "Add user authentication"

# Execute with guardrails
cco task --repo /path/to/repo --task "Add user authentication" --guarded

Architecture Overview

CCO implements a 3-tier memory architecture aligned with the paper:

Tier 1: Hot Memory (Always Loaded)

Constitution-level knowledge

AGENTS.md - Agent definitions and capabilities
CLAUDE.md - Project-specific instructions

Tier 2: Specialized Skills

19 domain-expert agents

Code roles: implementer, planner, reviewer, architect, debugger
Domain experts: security-researcher, database-expert, devops-engineer, etc.
Trigger patterns auto-select the right agent

Tier 3: Cold Memory (On-Demand)

Persistent knowledge via AOMA

.context/*.md - Architecture, decisions, known-issues
AOMA - SQLite-backed multi-project memory store
MCP tools for retrieval without token overhead

CLI Commands Overview

CCO provides 40+ commands organized into groups:

Group	Commands	Description
Core	`ask, version, help, doctor, plan, task, review, trace, cleanup`	Essential task execution commands
Task Management	`tasks start, finish, abandon, list`	Branch-based task lifecycle
Configuration	`config init, show, list, switch, delete, tier`	Profile and tier management
Memory	`mem add, ask, list, status, consolidate, export, sync`	AOMA long-term memory operations
Integration	`webhook, usage, secret, audit, obsidian, codex`	External system integration
Monitoring	`metrics, alert, alert-add`	Session metrics and alerting

MCP Server

CCO includes a Model Context Protocol server that exposes tools for external agents:

Tool	Description
`list_subsystems()`	Scan workspace for subsystems/components
`find_relevant_context()`	Search AOMA memory + .context files
`search_context_docs()`	Full-text search of context documents
`suggest_agent()`	Auto-select best agent for task
`memory_ingest/query/list/consolidate`	AOMA memory operations

python -m codified_orchestrator.mcp.server