CLI¶
Run arl COMMAND --help for every option.
| Command | Purpose |
|---|---|
arl init |
scaffold a target, a starter scenario and a CI workflow |
arl suites |
list built-in scenario packs |
arl run |
run a suite against a target (--repeats, --user-llm, --judge-llm, --junit-out, --results-out, --json-out, --min-pass-rate, --cassette/--record-mode, --otel) |
arl runs |
list past runs |
arl compare BASE NEW |
statistical comparison; exit code 1 on regression |
arl report RUN |
HTML report with step-by-step trace replay |
arl dashboard |
static dashboard across runs and versions |
arl ci-summary |
Markdown summary for PR comments |
arl redteam |
generate attack variants that breach the target |
arl label export / annotate / agreement |
human labels and judge agreement |
arl mcp |
serve the lab over MCP |
Global options: --version, --plugin MODULE (repeatable).
Record and replay¶
arl run ... --cassette c.json --record-mode record
arl run ... --cassette c.json --record-mode replay # offline, exact: agent, user, judge and tool calls