The Box

An agentic toolbox for Product Managers, designed to work with Claude Code.

It combines:

• Skills - Natural language interfaces for common PM tasks (e.g., data exploration, knowledge curation)
• CLI tools - Direct access to Datadog, Snowflake, Jira, Confluence, and Google Docs
• Knowledge base - Curated context that keeps Claude grounded in your domain, maintained organically via the /knowledge skill
• Self-maintaining - The toolbox improves itself as you use it

The tools run locally via ./box.sh, with some commands optionally using a Docker container for isolation.

Note: This repo is public, but the data/ directory is gitignored - it contains proprietary context specific to my work.

Sibling Directories

Code from external GitHub projects is stored in sibling directories under ../Code/:

../Code/
  acme/             # acme org's internal GitHub repositories
    acme-analytics  # acme analytics Github repository
    acme-core       # acme core Github repository
  beta/             # beta project Github repository 

Setup

cp env.example .env   # Add your credentials (Atlassian, Datadog, Snowflake)
./box.sh --setup      # Create venv and install dependencies

Prerequisites

• Python 3.11+: brew install python@3.11
• Docker: brew install docker-desktop
• Node.js (for MCP servers): brew install node

This toolbox has been tested on macOS environments.

Knowledge Base

Curated knowledge lives in _knowledge/ folders inside data/:

• data/_knowledge/ — cross-cutting reference (tool schemas, Slack directory, general domain context)
• data/<project>/_knowledge/ — initiative-specific knowledge (e.g., data/detection/_knowledge/)

The /knowledge skill owns the curation rules: file templates, when to update silently vs. ask first, and explicit review passes. Knowledge files are living documents updated in-place — only distilled insights from pulled documents or explorations belong here, not raw outputs.

Claude Skills

When using Claude Code, skills provide higher-level interfaces to the underlying libraries:

• /exploration - Data exploration: query user behavior (Datadog RUM), business metrics (Snowflake), and qualitative feedback (Jira). Intelligently routes to the right data source(s) based on your question. Outputs go in data/explorations/.
• /dd-admin - Investigate monitor triggering issues using Datadog's internal Monitor Admin APIs. Guided workflow for customer escalations.
• /knowledge - Knowledge base curation: review, update, and maintain _knowledge/ folders. Also runs organically during conversations to keep knowledge current.

MCP Servers

Claude Code also has direct API access via MCP servers, configured in .claude/settings.json:

• Datadog — search/analyze logs, monitors, dashboards, RUM events, incidents, services
• Atlassian — read/write Jira issues and Confluence pages
• Slack — search channels, read/send messages, manage canvases
• Snowflake — run queries, explore schemas, manage objects

Libraries

The libs/ directory contains the underlying CLI tools accessible via ./box.sh:

• Google (./box.sh google pull) - Pull Google Docs as local markdown via browser-based Apps Script
• Confluence (./box.sh confluence pull) - Pull Confluence pages as local markdown via REST API v2
• Jira (./box.sh jira) - Fetch Jira tickets with custom fields and ADF-to-text conversion
• Snowflake (./box.sh snowflake) - Execute SQL queries with INCLUDE directives and template variables
• Datadog (./box.sh datadog) - Query and aggregate RUM data, fetch session timelines, create/update notebooks for Product Analytics
• Metabase (./box.sh metabase) - Manage Metabase dashboards using YAML definitions with Terraform-like state tracking for pull/push workflows
• Analysis (./box.sh analysis) - Statistical analysis for CSV data: A/B test comparison and exploratory analysis with clustering
• DD Admin (./box.sh dd-admin) - Datadog internal admin tooling for debugging monitor evaluation
• Excalidraw (./box.sh draw) - Local diagram editor for architecture diagrams, wireframes, and concept sketches

Note: The DD Admin CLI requires Datadog enterprise credentials and VPN access to internal APIs. It will not work outside of Datadog's corporate network.

See libs/README.md for details, or run ./box.sh <command> --help.

Services

The services/ directory contains Docker services orchestrated via docker-compose:

• Excalidraw - Collaborative canvas with REST API running at http://localhost:3000

  • REST API for programmatic element creation/updates
  • Real-time sync between browser UI and API via WebSocket
  • Persistent state in data/_excalidraw/elements.json
  • ./box.sh draw - Start and open in browser
  • ./box.sh draw new <name> - Create new diagram file
  • ./box.sh draw list - List saved diagrams
  • See services/excalidraw/README.md for API documentation

• Graph Viewer - Interactive directed graph visualizer at http://localhost:5001

  • Upload CSV data (source, target, weight)
  • Filter by weight threshold
  • Multiple graph layouts (DAG, force-directed, circular, etc.)
  • Dark mode interface with dynamic edge width mapping
  • See services/graph-viewer/README.md for details