Notebooks, tasks, and workflows

Lattice separates interactive exploration, repeatable execution, and canonical workspace mutation.

Run a notebook

Open an .ipynb file. Markdown and existing outputs render immediately.

• Choose Run on one code cell or Run all in the notebook toolbar.
• Choose Cancel to interrupt the active run.
• The native desktop prefers an out-of-process ipykernel and falls back visibly when that environment is unavailable.
• Browser-oriented environments may use Pyodide with a bounded mounted-file bridge.

Successful outputs are merged back into the Jupyter document with a revision guard. The notebook remains readable if no kernel can start.

Run a task

A *.task/ package declares its runtime, project, command, inputs, outputs, and timeout in task.yaml. Open the package to inspect the manifest, then choose Run. The task surface streams stdout and stderr, shows status and duration, and links declared outputs. Cancel terminates an active task.

Tasks use local providers such as uv, Nix, or a configured system runtime. They should write declared outputs or return a proposal rather than modifying arbitrary workspace files invisibly.

Run a workflow

A *.workflow.yaml resource coordinates bounded task and proposal steps. The workflow surface shows whether automatic triggers are enabled, the raw YAML, step logs, recent executions, and any proposal produced by the run. Manual Run remains available when triggers are disabled.

Review a proposal

Open the proposal inbox from the activity rail. A proposal names its source, summary, affected paths, warnings, and semantic commands. Select the commands you want and apply them as one transaction, or reject the proposal without changing canonical files.

This review boundary is also how future agents and hosted automation can remain useful without receiving ambient write authority.