Build an Executive Analytics Agent with Claude Cowork, Bauplan MCP, Linear and GitHub Actions

Walkthrough of the executive analytics agent example project: an agent that answers analytical questions over a lakehouse and when the answer doesn't exist yet, files the work to build it as a permanent pipeline.The setup uses Claude Cowork as the interface, the Bauplan MCP for querying and validating against the lakehouse, Linear as the ticketing system and GitHub Actions to run Claude against the repo when an issue lands.Every question takes one of two paths:Path A: a gold table already answers it. The agent queries and returns the result. Nothing is written, nothing is reviewed.Path B: no table answers it yet, but the data is one query away. The agent runs the query, returns the result to the executive immediately, then files a Linear issue to turn that query into a permanent pipeline.Path B closes the loop automatically. Linear syncs the issue to GitHub, a GitHub Action runs Claude against the repository, the agent reads the issue and project docs, and validates the pipeline with a Bauplan dry run. An engineer reviews and merges, a scheduler materializes the gold table on main, and the next time that question is asked it hits path A.What's covered:

• Cloning the template repo and adding Anthropic and Bauplan API keys as GitHub Actions secrets
• Enabling GitHub Actions to create and approve pull requests
• Setting up the Claude Cowork project: instructions, repo context, Bauplan MCP as a custom connector, Linear MCP as a standard connector
• Linear setup: team, project, GitHub integration, and branch name format without forward slashes
• The project docs: CLAUDE.md, semantics.md, lakehouse.md (medallion architecture and pipeline rules), linear.md, the workflow doc, and the Claude YAML that drives the GitHub Action
• The TPC-H dataset used for the demo: customer, line item, nation, orders, parts, supplier, region
• A live run: prompting for line item count by customer, filing the Linear issue, watching the GitHub Action and the Bauplan job, and the resulting pipeline