tkmd Brings Deterministic Scheduling and Review Gates to AI Coding Agents

Summary

tkmd is a local, offline task and project manager for developers and AI coding agents, implemented as a Rust binary with the `tkmd` and `tk` commands. It stores tasks as Markdown cards with YAML frontmatter and uses Critical Path Method scheduling and Design Structure Matrices to calculate buffers, prioritize critical-path work, and detect dependency cycles. AI agents cannot move tasks directly to `done`: they must run a shell-based evaluation, submit the task to `pending-close`, and receive human sign-off, either directly or through an instructed proxy. The tool supports concurrent task claims with POSIX advisory file locking, automatic unblocking of downstream tasks, and resumption of an AI terminal session attached to a task. Users can operate it with an Obsidian Kanban setup, a standalone Markdown directory, or a repository-local backlog, while also viewing local and global scopes together. tkmd exposes a stdio MCP server compatible with tools including Claude Desktop, Codex, Gemini CLI, Cursor, Zed, OpenCode, and Pi, and ships an agent skill that reinforces its review workflow. State changes, evaluations, and sessions are recorded in append-only JSONL, which tkmd can analyze directly and users can query with DuckDB. The project offers an install script or Cargo installation and is dual-licensed under MIT and Apache 2.0.