Build Sprint

This lesson is a checklist, not a lecture. The building happens in your real workspace with Claude Code. Follow the phases, log your progress.

Phase 1: Setup (15 min)

• [ ] Create project directory
• [ ] Set up dependencies (package.json or requirements)
• [ ] Verify you have access to all required APIs/tools
• [ ] Create a basic test that proves connectivity (can you read from each data source?)

Phase 2: Core Logic (45 min)

• [ ] Implement the main agent loop or pipeline
• [ ] Connect to your data sources (MCP tools, APIs, files)
• [ ] Get the happy path working end to end
• [ ] Verify output for one real example

Phase 3: Error Handling (20 min)

• [ ] What happens when a tool call fails? Add retry or fallback.
• [ ] What happens with unexpected input? Add validation.
• [ ] What happens when the LLM returns unexpected output? Add parsing guards.
• [ ] Test each error path at least once.

Phase 4: Eval (20 min)

• [ ] Write your eval cases from the project brief as runnable tests
• [ ] Run all eval cases
• [ ] Fix any failures
• [ ] Run evals again to confirm

Phase 5: Polish (20 min)

• [ ] Add a README with setup and usage instructions
• [ ] Clean up code: remove debug logs, add essential comments
• [ ] If it's an MCP server: add it to your Claude Code config and test from Claude Code
• [ ] If it's a standalone tool: verify it runs with a single command

Build Log

Track your progress as you build. This becomes your retrospective material.