Test Command Reference

Full reference for hb test — the command that runs an adversarial test against your agent. Covers the test-category flag (which orchestrator to use), testing levels (unit / system / acceptance shortcuts), language and provider selection, the experiment name and description, behaviour flags (--wait, --fail-on), and the optional --endpoint override for testing against a different agent than the project's default.

hb test [OPTIONS]

Test Configuration:
  -t, --test-category     Test to run (default: humanbound/adversarial/owasp_agentic)
  --category              Shorthand alias for --test-category
  -l, --testing-level     Depth: unit | system | acceptance
  --deep                  Shortcut for --testing-level system
  --full                  Shortcut for --testing-level acceptance
  -n, --name              Experiment name (auto-generated if omitted)
  -d, --description       Experiment description
  --lang                  Language (default: english). Accepts codes: en, de, es
  -c, --context           Extra context for the judge (string or .txt file path)
  --provider-id           Provider to use (default: first available)

Behavior:
  --no-auto-start         Create without starting (manual mode)
  -w, --wait              Wait for completion
  --fail-on SEVERITY      Exit non-zero if the experiment produces
                          insights >= severity
                          Values: critical, high, medium, low, any

Endpoint Override (optional):
  -e, --endpoint          Agent integration config -- JSON string or file path.
                          Same shape as 'hb connect --endpoint'.
                          Overrides the project's default integration.

Note

The -e / --endpoint flag is only needed if your project was not connected with hb connect --endpoint, or if you want to temporarily test against a different agent. When a default integration is configured, hb test works with no additional flags. Your --endpoint JSON file can also include a telemetry section for white-box agentic testing -- see Agent Configuration File.

Reusing a previous run's configuration

Every experiment stores the configuration it ran with. Print it back and feed it straight into a new run:

hb experiments show <experiment-id> --config > bot-config.json
hb test --endpoint ./bot-config.json --wait

The printed JSON is the full configuration (integration, scope, context). hb test detects this and uses its integration block — scope and context always come from the project's current settings, not the file.