Ordewell is a task planner and execution system for AI coding agents that sits between a human goal and actual code changes. You describe what you want, the planner (Claude, Codex, or OpenCode in read-only mode) breaks it into an ordered list of tasks—each with a chosen model, runner, thinking level, and dependencies—and you can edit that plan before anything runs. Then it executes each task using the assigned runner, marking completion only when a specific marker appears in the output, not on exit code or model opinion alone. The source emphasizes that the plan is a visible, editable artifact (not hidden agent state), that different tasks get different models based on what they actually need, and that you can manually override verdicts or add tasks without losing prior work. Three interfaces ship built-in: a VS Code extension with a streaming timeline, a terminal UI over SSH, and a CLI for headless use. It integrates with existing subscriptions (Claude Code, Codex, OpenCode) and plugs into 25+ LLM providers via API key. The planner researches your repo read-only, asks clarifying questions inline, and refuses write operations during planning—mutation only happens in the runner phase. One open question: how does the "completion marker" get chosen or detected when the source doesn't explain that mechanism explicitly?