Middleware that gives the agent an explicit planning tool.
Contributes a write_todos tool plus a :todos state key. The agent
records and revises a task list as it works, which keeps a long multi-step
loop anchored to a plan instead of drifting — one of the cheapest quality
gains for multi-step agents. The current plan lives in graph state (and
therefore in checkpoints), so it survives pauses and resumes.
Encourage use from the agent's system prompt (e.g. "Before acting, write a todo list and keep it updated.").
Options
:description- override the tool description shown to the model
Summary
Functions
Builds a todo-list planning middleware. See the module doc for options.
Functions
@spec new(keyword()) :: LangEx.Middleware.t()
Builds a todo-list planning middleware. See the module doc for options.