CLI Overview¶

The genelab CLI is a thin dispatcher over registries and task configs. It does not own task logic; it discovers extensions, resolves a registered object, applies overrides, and calls the task or runner.

Command model¶

genelab [global options] <command> [arguments]

Area	Commands
Registry	`list robots\\|envs\\|tasks`, `info NAME`, `asset list\\|info\\|download\\|purge`
Runtime	`play TASK`, `train TASK`, `eval TASK CHECKPOINT`, `export TASK CHECKPOINT`, `benchmark --suite ...`
Project	`project new NAME`
Utilities	`cache`, `prof open`

Running genelab with no subcommand prints a landing page with quickstart commands and a count of registered robots / envs / tasks.

Global options¶

Option	Meaning
`--version`	Print the version and exit.
`--import MODULE`	Import an extension module before dispatch; repeatable.
`--no-entry-points`	Skip extensions from the `genelab.extensions` entry-point group.

Extension loading order¶

Every command that needs registry data loads extensions in this order:

Bundled asset zoo robots through load_bundled_asset_zoo().
Installed genelab.extensions entry points, unless --no-entry-points is set.
Repeated explicit --import MODULE values.

Use entry points for daily work and --import for local experiments.

Overrides¶

Runtime commands accept unknown --a.b.c VALUE options after the task id. The CLI forwards them to apply_overrides:

genelab play TASK_ID --env.simulation.dt 0.005

Use genelab info TASK_ID to list valid paths.

Interactive mode¶

When stdin is a TTY, the CLI can prompt for a missing task id, unknown registry name, invalid --agent, or unknown override path. In CI and scripts it raises the same errors directly.