CLI Development

Consult standards-for-python-development for the shared rules. Load python3-testing for test patterns.

Audience First

Before choosing presentation libraries, identify the primary consumer.

A CLI implemented inside an Agent Skill or plugin is agent-facing by default. Do not use Rich for its command output. Emit compact JSON on stdout, preferably from a Pydantic response model with model_dump_json(); do not indent it. Send diagnostics to stderr and keep the JSON schema stable.
Human-facing CLIs may use Typer + Rich for terminal presentation.
For mixed audiences, compact JSON is the automation contract; human formatting is an explicit presentation mode.

Agents must not need to parse tables, colours, progress bars, panels, or explanatory prose to consume a tool result.