wallfacer

About

wallfacer exists to keep AI-assisted engineering organized as projects grow, instead of scattering work across chat sessions. Ideas enter as chat, become versioned Markdown specs arranged in a dependency tree with a seven-state lifecycle, and dispatch as tasks onto a kanban board where each task runs as a host process in its own git worktree with auto-test, auto-retry, circuit breakers, and per-task token/cost budgets. A pluggable harness layer lets each task or agent role be pinned to Claude Code, Codex, Cursor, OpenCode, or Pi, and oversight surfaces provide event timelines, diffs, and AI-generated summaries for review. The system is locally run with no cloud dependency and notably has developed much of its own recent functionality. It targets developers who want spec-driven, parallel agent execution on their own machines.