Claude Orchestrator

A portfolio management system for Claude Code. Run multiple AI-assisted projects from a single parent directory — with shared agents, skills, automated daily reporting, and cross-project intelligence.

What It Does

• Manage multiple projects from one place — each gets its own Team Lead agent, tailored skills, and rules
• Daily morning digest — per-project intelligence reports at 8:45am, synthesised into one digest at 9:00am (with Gmail + Google Calendar integration)
• EOD wrap-up — compares what was planned vs what got done; surfaces tomorrow's priorities
• Nightly maintenance — syncs agent/skill registries, runs health checks, updates the task board
• Weekly entity scan — builds a cross-project map of people, companies, and systems
• Pre-session briefing — automatically injects today's context when you open any project
• Cross-project queries with /ask — ask anything about your portfolio in plain language
• Project archiving with /archive-project — clean offboarding with completion report

Requirements

• Claude Code CLI installed and authenticated (claude in your PATH)
• macOS or Linux (uses crontab for scheduling)
• Python 3 (for setup — usually pre-installed)
• Optional: Gmail and Google Calendar connected at claude.ai/settings/connectors for richer morning digests

Quick Start

1. Clone and run setup:

git clone https://github.com/YOUR_USERNAME/claude-orchestrator
cd claude-orchestrator
bash setup.sh

Setup asks for your name and timezone, configures settings.json, and registers all cron jobs automatically. Takes about 60 seconds.

2. Open Claude Code in this directory:

claude

3. Create your first project:

/project-scaffolder

The interactive wizard handles the rest — one question at a time.

How Projects Work

Each project lives in projects/{name}/ and gets:

• A CLAUDE.md, AGENTS.md, SKILLS.md, RULES.md, CHANGELOG.md, TASKS.md
• A Team Lead agent as the default point of contact
• Additional agents based on project type (dev, financial, business, research, misc)
• A morning report script that runs automatically at 8:45am

The orchestrator acts as the parent layer — it can see all projects, sync registries, and run cross-project workflows.

Each project can optionally use a different Claude account — the scaffolder will ask during setup. When a non-default account is configured, open that project with claude --profile <account> instead of claude.

Project Types

Agent Roster

Skills

Proprietary (Built-in)

Official (Anthropic)

Community (Vetted)

Automated Schedule

Times are in your local timezone. setup.sh converts them to UTC automatically when registering cron jobs.

To adjust the schedule after setup: crontab -e

Pre-Session Briefing

When you open Claude Code in any project for the first time each day, it automatically injects:

• Today's digest highlights (urgent items + agenda)
• Recent changelog entries (last 3 days)
• Open tasks for that project

This happens via a UserPromptSubmit hook in each project's .claude/settings.json. It fires once per calendar day — no overhead after that.

MCP Integration (Gmail + Google Calendar)

The morning digest and EOD wrap-up can pull from Gmail and Google Calendar for richer context:

1. Go to claude.ai/settings/connectors
2. Connect your Google account
3. Done — the MCP config is already at scripts/mcp-morning.json

Without this, digests still work — they just won't include email or calendar data.

Directory Structure

claude-orchestrator/
├── projects/                     # Your projects (gitignored — stays local)
├── reports/                      # Digests and logs (gitignored — stays local)
├── templates/
│   ├── agents/                   # Agent definition templates
│   └── project/                  # Project scaffold template
├── scripts/
│   ├── setup-morning-crons.sh    # Deploy morning scripts to all projects + update crontab
│   ├── morning-digest.sh         # 9:00am orchestrator digest
│   ├── eod-digest.sh             # 7:00pm EOD wrap-up
│   ├── daily-sync.sh             # 2:00am nightly maintenance
│   ├── weekly-entity-scan.sh     # Sunday entity map update
│   ├── pre-session-brief.sh      # Session briefing hook (called by project hooks)
│   └── templates/
│       └── morning-report.sh     # Per-project morning report template
├── .claude/
│   ├── agents/                   # Orchestrator-level agent definition
│   └── skills/                   # project-scaffolder, ask, archive-project, registry-sync
├── AGENTS.md                     # Master agent registry
├── SKILLS.md                     # Master skills registry
├── WORKFLOWS.md                  # Scheduled workflow definitions
├── RULES.md                      # Shared guardrails (inherited by all projects)
├── TASKS.md                      # Cross-project task board
├── ENTITIES.md                   # Cross-project entity map (auto-generated weekly)
├── settings.json                 # Your personal config (gitignored — created by setup.sh)
├── settings.example.json         # Config template (committed)
└── setup.sh                      # First-time setup script

Adding a New Project

/project-scaffolder

After scaffolding, register its morning cron:

bash scripts/setup-morning-crons.sh

Safe to re-run at any time — never duplicates crontab entries.

Ingesting an Existing Project

The scaffolder supports ingesting existing projects of any type — code repos, financial models, research archives, business documents:

/project-scaffolder
> 2  (Existing project)
> /path/to/your/project

It analyses the directory, infers the project type, and layers in the orchestrator format non-destructively. Your existing files are untouched.

Bulk ingestion (option 3 in the wizard) ingests multiple projects in one pass and opens a tmux tab per project.

Running Digests Manually

Any automated script can be triggered manually:

bash scripts/morning-digest.sh
bash scripts/eod-digest.sh
bash scripts/daily-sync.sh
bash scripts/weekly-entity-scan.sh

FAQ

Do I need to be online for cron jobs to run? Yes — the scripts call the claude CLI, which requires an active internet connection and a valid Claude Code session. Failed runs are logged to reports/cron-run.log.

Where are my reports? reports/daily/{YYYY-MM-DD}/ — one directory per day. DAILY-DIGEST.md is the morning synthesis, EOD-DIGEST.md is the evening wrap-up.

How do I disable a cron job?

crontab -e

Comment out or delete the relevant line.

My project is done. How do I archive it?

/archive-project my-project-name

Generates a completion report, moves the project to projects/archived/, removes its crontab entry, and updates all registries.

Can I use this without the cron jobs? Yes — everything works on-demand. The scheduling is optional. Open Claude Code, use /project-scaffolder to create projects, and interact normally. The crons just add automated daily intelligence on top.

Customising Rules

RULES.md defines guardrails inherited by all projects. Edit it to change global defaults. Individual projects can add a local RULES.md to layer in project-specific rules — the stricter rule always wins.

License

MIT