CLI Commands

Top-Level Commands

Command	Purpose
robosmith run	Run the full natural-language-to-policy pipeline.
robosmith resume	Resume an interrupted pipeline run from checkpoint.json.
robosmith envs	List and filter registered simulation environments.
robosmith trainers	Show available trainer backends and algorithms.
robosmith deps	Show installed and missing optional dependencies.
robosmith config	Print default ForgeConfig as JSON.
robosmith version	Print package version.
robosmith inspect	Inspect datasets, environments, policies, robots, and compatibility.
robosmith diag	Analyze or compare trajectory rollouts.
robosmith gen	Generate wrapper code.
robosmith auto	Run compound agentic workflows.
robosmith runs	List, inspect, compare, and clean run directories.

robosmith run

robosmith run --task "A Franka arm picks up a red cube"

Flag	Default	Description
--task, -t	required	Natural language task description.
--robot, -r	arm	Robot type: arm, quadruped, biped, dexterous_hand, mobile_base, custom.
--model, -m	none	Specific robot model, for example franka, unitree_go2, shadow_hand.
--algo, -a	auto	Algorithm: ppo, sac, td3, auto.
--time-budget	60	Max training time in minutes.
--num-envs	1024	Number of parallel environments requested.
--push-to-hub	none	HuggingFace repo ID for pushed artifacts.
--candidates, -c	4	Reward candidates per iteration.
--skip, -s	none	Stages to skip: scout, intake, delivery.
--backend, -b	auto	Training backend: sb3, cleanrl, or another registered backend.
--llm, -L	auto	Provider or exact LiteLLM model string.
--scout	semantic_scholar	Search backend: semantic_scholar, arxiv, both.
--config	auto-detect	Path to robosmith.yaml.
--dry-run	false	Parse and plan only; do not train.
--verbose, -v	false	Write debug logs to robosmith_runs/latest.log.

robosmith resume

robosmith resume run_20260415_182058_a64796
robosmith resume run_20260415 --runs-dir ./robosmith_runs --verbose

Argument or flag	Description
run_id	Full run ID or unique prefix.
--runs-dir	Base directory containing run_* directories.
--verbose, -v	Debug logging.

robosmith envs

robosmith envs --robot arm --framework gym --tags "pick place"

Flag	Description
--robot, -r	Filter by robot type.
--framework, -f	Filter by framework.
--tags, -t	Comma-separated tags to search.
--env-type, -e	Filter by environment type.

robosmith inspect

robosmith inspect dataset lerobot/aloha_mobile_cabinet --schema --quality
robosmith inspect env Ant-v5 --obs-docs --sample
robosmith inspect policy lerobot/smolvla_base
robosmith inspect robot path/to/robot.urdf
robosmith inspect compat POLICY TARGET --fix

Subcommand	Extra flags
dataset IDENTIFIER	--json, --schema, --quality, --sample N
env IDENTIFIER	--json, --obs-docs, --sample
policy IDENTIFIER	--json, --config, --requirements
robot IDENTIFIER	--json
compat A B [C]	--json, --fix

robosmith diag

robosmith diag trajectory path/to/rollout.hdf5
robosmith diag compare rollout_a.hdf5 rollout_b.hdf5 --json

Subcommand	Description
trajectory PATH	Analyze one trajectory set.
compare PATH_A PATH_B	Compare two trajectory sets.

Both support --json.

robosmith gen

robosmith gen wrapper POLICY TARGET
robosmith gen wrapper POLICY TARGET --no-llm -o adapter.py

Flag	Description
--output, -o	Write generated code to a file.
--no-llm	Use template-based generation.

robosmith auto

robosmith auto integrate POLICY TARGET --verbose -o adapter.py

Flag	Description
--output, -o	Write generated wrapper to a file.
--json, -j	Output machine-readable final state.
--verbose, -v	Show step-by-step execution.

robosmith runs

robosmith runs list --limit 20
robosmith runs inspect run_20260415 --log --reward
robosmith runs compare run_20260415 run_20260416
robosmith runs clean --older-than 14 --dry-run

Subcommand	Important flags
list	--runs-dir, --status, --limit
inspect RUN_ID	--runs-dir, --log, --reward
compare RUN_A RUN_B	--runs-dir
clean	--runs-dir, --older-than, --dry-run, --yes