Sprint 1 - viz-platform Plugin Lessons Learned

Context

Implemented the viz-platform plugin for Claude Code marketplace, providing Dash Mantine Components validation, Plotly chart creation, and design token theming.

Completed Work

13 issues closed (100% completion)
MCP server with 4 tool modules (config, component_registry, dmc_tools, chart_tools, theme_tools)
7 commands, 3 agents, full documentation
94 tests passing with 68-99% coverage

Key Lessons

1. Test Assertions Must Match Actual API

Problem: 4 tests failed because assertions used wrong keys (active vs active_theme, missing vs missing_required)

Solution: Read implementation code before writing test assertions, or run tests early to catch mismatches

Prevention: Create tests incrementally alongside implementation, not all at once at the end

2. Never Quote Effort Labels as Time Estimates

Problem: Stated tests would take "3-5 days" based on Gitea's Effort/L label - they took 10 minutes

Solution: Labels indicate relative complexity for sprint planning, not actual duration

Prevention: Follow instruction "Never give time estimates" - focus on what needs doing, not how long

3. Large MCP Responses Are Normal

Problem: "Output too large (100KB)" message confused user who thought it was a network error

Solution: This is Claude Code saving large outputs to files - normal behavior, not an error

Prevention: When encountering large response messages, explain they're file saves, not errors

Metrics

Issues: 13/13 closed
Tests: 94/94 passing
Coverage: config 82%, registry 92%, dmc_tools 88%, chart_tools 68%, theme_tools 99%

Tags: viz-platform, testing, mcp-server, sprint-complete