DeadWaveWave/waveforge

Implements an MCP (Model Context Protocol) server that stores all task data, logs, and metadata as versioned plain-text files in a local `.wave` directory, enabling Git-based collaboration and full data provenance. Features a two-level task hierarchy (overall plans and specific steps) with automated workflow progression—completing the final step auto-advances to the next plan—plus a three-tier hint system delivering contextual guidance to AI agents at task, plan, and step levels. Integrates with AI-first IDEs (Cursor, VSCode, Kiro) via stdio transport and uses file locking with state versioning to maintain consistency across parallel multi-agent operations.