Commands Reference

Global Options

These options are available on all commands.

Flag	Description	Default
`-s, --skills-dir <dir>`	Override skills directory (repeatable)	-
`-c, --config <path>`	Path to config file (default: auto-detect dojo.toml)	-
`-d, --cwd <dir>`	Working directory for config and skill discovery	-
`-v, --version`	Output the current version	-
`-h, --help`	Display help for command	-

Run evals against discovered skills. Results are saved as JSON reports per-skill.

dojo run [skill] [options]

Argument	Description	Required
`skill`	Filter by skill name (supports glob patterns)	No

Flag	Description	Default
`-e, --eval <name>`	Run only evals matching this name	-
`-V, --variant <name>`	Run only a specific variant (by name)	-
`-p, --parallelism <n>`	Max concurrent eval runs	CPU cores
`--no-parallelism`	Run evals sequentially	-
`-o, --output <path>`	Write combined report to file	-
`-i, --inspect`	Show key session events (model, prompt, tool calls, errors). Full event stream is written to logs.json.	-
`-t, --eval-type <type>`	Filter by eval type: “selection”, “effectiveness”, or “all”	all
`--selection`	Run only selection evals (shortcut for -t selection)	-
`--effectiveness`	Run only effectiveness evals (shortcut for -t effectiveness)	-
`-m, --evaluation-model <model>`	Override evaluation model	-
`-j, --judge-model <model>`	Override judge model	-
`--model-provider <provider>`	Override model provider	-
`-f, --fixture <name>`	Filter effectiveness evals to a specific fixture	-
`--judge-filter <id>`	Filter to a specific judge (format: “provider/model”)	-
`--keep-sandbox`	Don’t clean up sandbox temp directories after run	-
`-y, --yes`	Skip confirmation prompts for large eval runs	-

List all discovered skills and evals in a tabular format.

dojo list [options]

Validate the config, skills, and evals. Reports errors without running anything.

dojo validate [options]

v0.6.1