Cortex is an 11-agent productivity system that turns Claude Code into a PM operating system. It orchestrates specialized agents to automate meeting prep, triage, intelligence gathering, stakeholder communication, and cross-functional coordination.
This is a template repository. Fork it, fill in your context, connect your tools, and start running workflows.
Product leaders (Directors, VPs, Senior PMs) who:
- Already use Claude Code as a daily tool
- Understand MCP basics (servers, tools, configuration)
- Manage a team of PMs or own a multi-product portfolio
- Want to automate the repetitive parts of PM leadership
This is not a tutorial on Claude Code or MCP. If you need that, start with the Claude Code docs first.
11 specialized agents, each with full instructions and quick-reference prompts:
| # | Agent | What It Does |
|---|---|---|
| 01 | Strategy & Roadmap | Prioritization, roadmap planning, scenario analysis |
| 02 | Team Leadership | 1:1 prep, feedback drafting, team capacity tracking |
| 03 | Stakeholder Comms | Executive updates, talking points, visibility strategy |
| 04 | Customer Insights | Customer evidence synthesis, win/loss analysis, segment patterns |
| 05 | PM Operations | PRD review, backlog health, process checks |
| 06 | Launch & GTM | Launch plans, readiness checks, sales enablement |
| 07 | Personal Assistant | Morning triage, meeting prep, weekly review, delegation tracking |
| 08 | Competitive Intel | Competitor briefs, threat assessment, analyst report synthesis |
| 09 | Feature Context | Feature status, design alignment, cross-product coordination |
| 10 | Engineering Partnership | Capacity asks, PI planning, tech debt negotiation |
| 11 | Metrics & Analytics | Metrics tracking, anomaly detection, trend analysis |
Pre-built frameworks: RICE scoring, MoSCoW, Impact-Effort matrix, Kano model, launch tiers, SBI feedback, and more.
Parallel workflows: Morning routine (8 agents), weekly review (11 agents), 1:1 prep (4 agents), intelligence sweeps, end-of-day capture. Claude Code's Task tool spawns subagents in parallel, reducing 15-20 minute workflows to 2-3 minutes.
Shared-state communication: Agents communicate across sessions through structured files: logs, status snapshots, inboxes, caches, insights, and daily/weekly summaries.
MCP integration layer: Pre-configured patterns for Notion, Slack, Google Calendar, Gong, and competitive intelligence tools.
Files throughout this template use {{PLACEHOLDER_NAME}} (double-curly-brace syntax) to mark values you need to customize. For example:
| **Name** | Chris Mayfield |
| **Title** | VP of Product |
| **Company** | {{COMPANY_NAME}} |
See CUSTOMIZATION.md for the full placeholder reference and priority order for filling them in.
- Fork or clone this repository
- Fill out
skills/context/config.mdwith your name, title, company, team, products, and tool configuration. This is the single most important file. - Define your team in
skills/context/team-roster.md - Connect MCP tools (Notion, Slack, Calendar at minimum)
- Run
Triageto test basic flow - Run
Run: morning-routinefor full parallel workflow
See SETUP.md for the complete step-by-step guide.
CLAUDE.md is the entry point. When you open Claude Code in this repo, it reads CLAUDE.md and acts as the orchestrator. Based on your request, it:
- Routes to the right agent(s) using the dispatch table
- Loads agent instructions for context and available modes
- Reads/writes shared-state files for cross-session continuity
- Queries external tools via MCP for live data
- Synthesizes results and presents a unified output
For the full system design, see ARCHITECTURE.md.
| File | Purpose |
|---|---|
| ARCHITECTURE.md | System concepts, design decisions, data flow |
| SETUP.md | Step-by-step installation and configuration |
| CUSTOMIZATION.md | What to change, placeholder reference, per-file guidance |
| Say This | What Happens |
|---|---|
| "Triage" | Personal Assistant scans comms, flags priorities |
| "Run: morning-routine" | 8 agents run in parallel for daily situational awareness |
| "Run: eod-sweep" | Captures day's outputs, action items, pending threads |
| "Week ahead" | Monday planning with calendar and priorities |
| "Run: weekly-review" | 11 agents for comprehensive Friday wrap-up |
| "Prep for [meeting]" | Meeting-specific context and talking points |
| "Run: 1on1-prep [name]" | Deep 1:1 prep across 4 agents |
| "Run: prioritize [topic]" | Multi-framework prioritization analysis |
cortex/
├── CLAUDE.md # Orchestrator entry point
├── agents/ # 11 agent definitions
│ └── [NN-agent-name]/
│ ├── instructions.md # Full instructions + modes
│ └── prompts.md # Quick reference
├── skills/ # Shared knowledge
│ ├── infrastructure/ # System behavior (error handling, token efficiency, etc.)
│ ├── frameworks/ # PM frameworks (RICE, MoSCoW, Kano, etc.)
│ ├── context/ # Your team, products, competitors, config
│ ├── orchestration/ # Parallel workflow definitions and routing
│ ├── brand-guidelines/ # Brand and style guidance
│ └── templates/ # Output formats, log entries, reusable blocks
├── shared-state/ # Cross-agent communication
│ ├── logs/ # Append-only (handoffs, decisions, accomplishments)
│ ├── status/ # Current state snapshots
│ ├── inbox/ # Per-agent message queues
│ ├── cache/ # MCP query result caches
│ ├── insights/ # Intelligence findings
│ ├── daily/ # Daily summaries
│ ├── weekly/ # Weekly summaries
│ ├── research/ # Research briefs
│ └── archive/ # Historical data
└── attachments/ # Binary files, diagrams, specs