CLI Reference

Browse agentbench commands by category. Search by command, argument, or description; select a command to inspect its usage and parameters. JSON API commands are listed in a separate category.

Search commandsCommand group

32 of 32 commands

Command Description

GeneralExecution, evaluation, recovery, and result management.

Run all enabled Agents whose status is ready.

agentbench run

agentbench run [-h] [--sdk NAME] [--sdk-options PATH]
                      [--case-retries CASE_RETRIES]
                      [--retry-delay RETRY_DELAY] [--agent ID] [-y]
                      [--env-file PATH] [--results-dir DIR | --output PATH]
                      [--no-view] [--model MODEL]
                      [--llm-trace-max-bytes BYTES]

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`--sdk`Option	NAME	No	—	Select an adapter directory listed by 'agentbench sdk list'. Omit only when exactly one adapter is available.
`--sdk-options`Option	PATH	No	—	JSON object of options passed to the selected SDK.
`--case-retries`Option	int	No	—	Maximum extra attempts for safe transient Case failures; 0 disables them.
`--retry-delay`Option	float	No	—	Initial Case retry backoff in seconds.
`--agent`Option	ID	No	[]	Run only this enabled ready Agent; repeat to select several.
`-y, --yes`Option	flag	No	false	Confirm this execution without prompting.
`--env-file`Option	PATH	No	—	Load host secrets and defaults from PATH instead of .env.
`--results-dir`Option	DIR	No	—	ABB result directory; managed Suites save DIR/suites/<suite-id>/events.json. Created if missing; independent of SDK output.
`--output`Option	PATH	No	—	Deprecated result location. For managed Suites, a file path selects its parent directory; the requested filename is not created. Use --results-dir instead.
`--no-view`Option	flag	No	false	Save results without starting the local live viewer.
`--model`Option	MODEL	No	—	Target model ID; defaults to the selected provider model environment variable.
`--llm-trace-max-bytes`Option	BYTES	No	262144	Legacy option: streaming memory spool threshold; payloads are never truncated.

Terminal help: agentbench run --help

Open a saved benchmark result in the local viewer.

agentbench view

agentbench view [-h] [--host HOST] [--port PORT] result_log

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`result_log`Positional	string	Yes	—	Path to an AgentBench JSON result snapshot.
`--host`Option	string	No	127.0.0.1	—
`--port`Option	viewer_port	No	8765	—

Terminal help: agentbench view --help

Run one adapting Agent and promote it to ready after execution succeeds.

agentbench certify

agentbench certify [-h] [-y] [--registry REGISTRY] [--sdk NAME]
                          [--sdk-options PATH] [--case-retries CASE_RETRIES]
                          [--retry-delay RETRY_DELAY] [--no-view]
                          [--env-file PATH]
                          [--results-dir DIR | --output PATH] [--model MODEL]
                          [--llm-trace-max-bytes BYTES]
                          agent_id

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`-y, --yes`Option	flag	No	false	Confirm this execution without prompting.
`--registry`Option	Path	No	<checkout>/resources/registry.toml	Agent registry path
`--sdk`Option	NAME	No	—	Select an adapter directory listed by 'agentbench sdk list'. Omit only when exactly one adapter is available.
`--sdk-options`Option	PATH	No	—	JSON object of options passed to the selected SDK.
`--case-retries`Option	int	No	—	Maximum extra attempts for safe transient Case failures; 0 disables them.
`--retry-delay`Option	float	No	—	Initial Case retry backoff in seconds.
`--no-view`Option	flag	No	false	Save results without starting the live viewer.
`agent_id`Positional	string	Yes	—	Registered adapting Agent to certify.
`--env-file`Option	PATH	No	—	Load host secrets and defaults from PATH instead of .env.
`--results-dir`Option	DIR	No	—	ABB result directory; managed Suites save DIR/suites/<suite-id>/events.json. Created if missing; independent of SDK output.
`--output`Option	PATH	No	—	Deprecated result location. For managed Suites, a file path selects its parent directory; the requested filename is not created. Use --results-dir instead.
`--model`Option	MODEL	No	—	Target model ID; defaults to the selected provider model environment variable.
`--llm-trace-max-bytes`Option	BYTES	No	262144	Legacy option: streaming memory spool threshold; payloads are never truncated.

Terminal help: agentbench certify --help

Run one enabled Agent and save traces

agentbench observe

agentbench observe [-h] [--agent AGENT] [--registry REGISTRY] [--list]
                          [--input INPUT] [--output OUTPUT]
                          [--env-file ENV_FILE] [--model MODEL]
                          [--timeout TIMEOUT] [--show SHOW]
                          [AGENT]

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`selection`Positional	AGENT	No	—	Enabled Agent number or ID; omit to choose interactively
`--agent`Option	string	No	—	Alias for the positional Agent number or ID
`--registry`Option	Path	No	<checkout>/resources/registry.toml	—
`--list`Option	flag	No	false	Show all enabled Agents and exit
`--input`Option	Path	No	—	Native input JSON file
`--output`Option	Path	No	<path>	—
`--env-file`Option	Path	No	—	—
`--model`Option	MODEL	No	—	Target model name (not an Agent number); defaults to the selected provider model
`--timeout`Option	float	No	—	Positive execution timeout in seconds
`--show`Option	Path	No	—	Review a saved observe run directory offline

Terminal help: agentbench observe --help

Evaluate one Agent on independent SDK Cases

agentbench evaluate

agentbench evaluate [-h] [-y] [--registry REGISTRY]
                           [--env-file ENV_FILE] [--model MODEL] [--sdk NAME]
                           [--sdk-options PATH] [--case-retries CASE_RETRIES]
                           [--retry-delay RETRY_DELAY] [--no-view]
                           [--llm-trace-max-bytes LLM_TRACE_MAX_BYTES]
                           [--results-dir DIR | --result-output PATH]
                           [--output OUTPUT] [--timeout TIMEOUT]
                           [--cases CASES] [--max-steps MAX_STEPS]
                           [--case-file CASE_FILE] [--judge JUDGE]
                           [selection]

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`-y, --yes`Option	flag	No	false	Confirm this execution without prompting.
`selection`Positional	string	No	—	Enabled Agent number or ID
`--registry`Option	Path	No	<checkout>/resources/registry.toml	—
`--env-file`Option	Path	No	—	—
`--model`Option	model_name	No	—	—
`--sdk`Option	NAME	No	—	Select an adapter directory listed by 'agentbench sdk list'. Omit only when exactly one adapter is available.
`--sdk-options`Option	PATH	No	—	JSON object of options passed to the selected SDK.
`--case-retries`Option	int	No	—	Maximum extra attempts for safe transient Case failures; 0 disables them.
`--retry-delay`Option	float	No	—	Initial Case retry backoff in seconds.
`--no-view`Option	flag	No	false	Save results without starting the live viewer.
`--llm-trace-max-bytes`Option	int	No	262144	—
`--results-dir`Option	DIR	No	—	ABB result directory; managed Suites save DIR/suites/<suite-id>/events.json. Created if missing; independent of SDK output.
`--result-output`Option	PATH	No	—	Deprecated result location. For managed Suites, a file path selects its parent directory; the requested filename is not created. Use --results-dir instead.
`--output`Option	Path	No	—	Override the SDK output option
`--timeout`Option	float	No	—	Override the SDK timeout option (seconds)
`--cases`Option	int	No	—	Number of independent Cases (default: Registry case count)
`--max-steps`Option	int	No	—	Override Registry step: SDK upper bound on dialogue steps per Case
`--case-file`Option	Path	No	—	Import abb.cases.v1 JSON with --sdk native or local
`--judge`Option	string	No	—	With --sdk native/local: local, kuma or a trusted module:object; omit for execution only

Terminal help: agentbench evaluate --help

Archive unreferenced local result history.

agentbench clean

agentbench clean [-h] [--dry-run] [-y]

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`--dry-run`Option	flag	No	false	List history without changing files.
`-y, --yes`Option	flag	No	false	Confirm cleanup without prompting; stop active runs first.

Terminal help: agentbench clean --help

Continue unfinished work in a saved Suite.

agentbench resume

agentbench resume [-h] [--suite-root SUITE_ROOT] [--env-file ENV_FILE]
                         suite

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`suite`Positional	string	Yes	—	Saved Suite ID, directory, or events.json path.
`--suite-root`Option	string	No	—	Directory containing saved Suites; defaults to results/suites.
`--env-file`Option	string	No	—	Credential/default file; secrets are never read from saved plans.

Terminal help: agentbench resume --help

Retry one unfinished Case using its original inputs.

agentbench retry

agentbench retry [-h] [--suite-root SUITE_ROOT] [--env-file ENV_FILE]
                        --agent AGENT --case CASE_NUMBER
                        suite

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`suite`Positional	string	Yes	—	Saved Suite ID, directory, or events.json path.
`--suite-root`Option	string	No	—	Directory containing saved Suites; defaults to results/suites.
`--env-file`Option	string	No	—	Credential/default file; secrets are never read from saved plans.
`--agent`Option	string	Yes	—	Exact Agent ID from the original Suite.
`--case`Option	int	Yes	—	One-based Case number.

Terminal help: agentbench retry --help

Rerun a saved Case or Suite with fresh Agent execution and judging.

agentbench reuse

agentbench reuse [-h] [--suite-root SUITE_ROOT] [--agent AGENT]
                        [--case CASE_NUMBER] [--output-root OUTPUT_ROOT]
                        [--env-file ENV_FILE] [--model MODEL]
                        [--max-steps MAX_STEPS]
                        suite

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`suite`Positional	string	Yes	—	Suite ID/path (all Cases), or Case ID, artifact run ID, saved Case/attempt path (one Case).
`--suite-root`Option	string	No	—	Directory to search for saved Suites; defaults to project results and indexed Suites.
`--agent`Option	string	No	—	Exact Agent ID; use with --case and a Suite source.
`--case`Option	int	No	—	One-based Case number; use with --agent.
`--output-root`Option	string	No	—	Only join/create reuse Suites under this directory; new batches otherwise use the original parent.
`--env-file`Option	string	No	—	Current credential/default file; secrets are never copied from a plan.
`--model`Option	string	No	—	Agent model for the new Suite; defaults to the saved model.
`--max-steps`Option	int	No	—	Positive SDK max_steps option for the new Suite.

Terminal help: agentbench reuse --help

Match saved Suite executions against Agent ground truth.

agentbench gt

agentbench gt [-h] --suite SUITE [--env-file ENV_FILE]

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`--suite`Option	Path	Yes	—	—
`--env-file`Option	string	No	—	—

Terminal help: agentbench gt --help

agentImport and inspect Agent source for onboarding.

Import an Agent and list files for onboarding.

agentbench agent add

agentbench agent add [-h] [--agents-dir DIRECTORY] [-b] [-c]
                            [--registry REGISTRY]
                            [--build-settings BUILD_SETTINGS]
                            [--build-model BUILD_MODEL] [--answers ANSWERS]
                            [--with-observe] [--agent-timeout AGENT_TIMEOUT]
                            [--adapter-context ADAPTER_CONTEXT]
                            [--env-file ENV_FILE] [--model MODEL]
                            [--output OUTPUT] [--no-view] [-y] [--sdk NAME]
                            [--sdk-options PATH]
                            SOURCE

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`repository`Positional	SOURCE	Yes	—	HTTPS GitHub repository URL or absolute local directory.
`--agents-dir`Option	DIRECTORY	No	<checkout>/resources/agents	Parent of numbered Agent folders (default: resources/agents).
`-b, --build`Option	flag	No	false	Generate, validate and save files one at a time; resume valid files.
`-c, --certify`Option	flag	No	false	Run existing certification using generated or manual files.
`--registry`Option	Path	No	<checkout>/resources/registry.toml	—
`--build-settings`Option	Path	No	—	TOML [build] overrides for generation.
`--build-model`Option	string	No	—	OpenRouter model for configuration generation only.
`--answers`Option	Path	No	—	Text answers to a previous build plan's questions.
`--with-observe`Option	flag	No	false	With -b, generate Observe fields from the input definition.
`--agent-timeout`Option	float	No	—	With -b, override the generated Agent timeout (default: 300 seconds).
`--adapter-context`Option	Path	No	—	With -b, explicit deployment overrides as a JSON object; omitted by default.
`--env-file`Option	Path	No	—	Host environment file; never included in source context.
`--model`Option	string	No	—	Target provider model for certification, independent of --build-model.
`--output`Option	Path	No	—	Certification result path.
`--no-view`Option	flag	No	false	Certify without starting the live viewer.
`-y, --yes`Option	flag	No	false	Confirm certification without prompting.
`--sdk`Option	NAME	No	—	Select an adapter directory listed by 'agentbench sdk list'. Omit only when exactly one adapter is available.
`--sdk-options`Option	PATH	No	—	JSON object of options passed to the selected SDK.

Terminal help: agentbench agent add --help

sdkList and inspect evaluation SDK plugins.

List evaluation adapters found in the SDK directory.

agentbench sdk list

agentbench sdk list [-h]

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit

Terminal help: agentbench sdk list --help

Load one SDK and show its execution interface.

agentbench sdk show

agentbench sdk show [-h] name

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`name`Positional	string	Yes	—	SDK adapter directory name.

Terminal help: agentbench sdk show --help

trajectoryBuild or show the derived step sequence of a saved execution.

Derive trajectory.json from the raw evidence of one attempt directory

agentbench trajectory build

agentbench trajectory build [-h] [--write] [--force] directory

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`directory`Positional	Path	Yes	—	Attempt directory (run.json or evaluation/ inside)
`--write`Option	flag	No	false	Save trajectory.json beside the evidence instead of printing it
`--force`Option	flag	No	false	Replace an existing trajectory.json whose content differs

Terminal help: agentbench trajectory build --help

Render the saved or derived trajectory as step text

agentbench trajectory show

agentbench trajectory show [-h]
                                  [--mode {full,truncate_middle,response_only}]
                                  directory

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`directory`Positional	Path	Yes	—	—
`--mode`Option	string	No	truncate_middle	—Choices: full, truncate_middle, response_only

Terminal help: agentbench trajectory show --help

exp1Run behavioural failure detection and report Detection@k

Execute and assess Cases in order, stopping at the first confirmed hit

agentbench exp1 run

agentbench exp1 run [-h] --issue ISSUE --cases CASES
                           [--results-dir RESULTS_DIR]
                           [--prompt-template PROMPT_TEMPLATE]
                           [--scenario-file SCENARIO_FILE] [--model MODEL]
                           [--timeout TIMEOUT] [--root ROOT]
                           [--env-file ENV_FILE]
                           agent_id

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`agent_id`Positional	string	Yes	—	—
`--issue`Option	int	Yes	—	Ground-truth Issue number
`--cases`Option	Path	Yes	—	Absolute path to an abb.cases.v1 JSON file (1-50 Cases)
`--results-dir`Option	Path	No	—	—
`--prompt-template`Option	Path	No	—	Record the shared testing prompt template
`--scenario-file`Option	Path	No	—	Record this Agent's shared scenario
`--model`Option	string	No	—	Target model override; oracle uses the shared provider default model
`--timeout`Option	float	No	2400	—
`--root`Option	Path	No	<checkout>	—
`--env-file`Option	Path	No	—	—

Terminal help: agentbench exp1 run --help

Assess saved trajectories without executing Cases

agentbench exp1 match

agentbench exp1 match [-h] --experiment EXPERIMENT [--retry-failed]
                             [--root ROOT] [--env-file ENV_FILE]

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`--experiment`Option	Path	Yes	—	—
`--retry-failed`Option	flag	No	false	Explicitly retry failed/interrupted oracle calls
`--root`Option	Path	No	<checkout>	—
`--env-file`Option	Path	No	—	—

Terminal help: agentbench exp1 match --help

Read saved Detection@k results

agentbench exp1 result

agentbench exp1 result [-h] --experiment EXPERIMENT [--root ROOT]
                              [--env-file ENV_FILE]

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`--experiment`Option	Path	Yes	—	Saved experiment directory; repeat to aggregate distinct Issues
`--root`Option	Path	No	<checkout>	—
`--env-file`Option	Path	No	—	—

Terminal help: agentbench exp1 result --help

exp3Export and validate Issue repair projects

Export pinned source/ and only the Issue object

agentbench exp3 genProject

agentbench exp3 genProject [-h] [--noRun] --issue ISSUE --output OUTPUT
                                  [--records-dir RECORDS_DIR] [--root ROOT]
                                  agent_id

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`agent_id`Positional	string	Yes	—	—
`--noRun`Option	flag	No	false	Export source and Issue without preparing a runnable Agent
`--issue`Option	int	Yes	—	—
`--output`Option	Path	Yes	—	New absolute directory outside ABB
`--records-dir`Option	Path	No	—	Host provenance directory outside the exported project
`--root`Option	Path	No	<checkout>	—

Terminal help: agentbench exp3 genProject --help

Replay the original GT Case against frozen external source

agentbench exp3 run

agentbench exp3 run [-h] --issue ISSUE --source SOURCE [--runs RUNS]
                           [--results-dir RESULTS_DIR] [--model MODEL]
                           [--timeout TIMEOUT] [--env-file ENV_FILE]
                           [--root ROOT]
                           agent_id

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`agent_id`Positional	string	Yes	—	—
`--issue`Option	int	Yes	—	—
`--source`Option	Path	Yes	—	Absolute path to the modified source/ directory
`--runs`Option	int	No	3	Independent repair executions (default: 3)
`--results-dir`Option	Path	No	—	—
`--model`Option	string	No	—	Target model override (execution only)
`--timeout`Option	float	No	2400	Timeout per Case execution in seconds
`--env-file`Option	Path	No	—	—
`--root`Option	Path	No	<checkout>	—

Terminal help: agentbench exp3 run --help

Read a saved repair result without executing anything

agentbench exp3 result

agentbench exp3 result [-h] --experiment EXPERIMENT [--root ROOT]

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`--experiment`Option	Path	Yes	—	—
`--root`Option	Path	No	<checkout>	—

Terminal help: agentbench exp3 result --help

exp3 judgeJudge saved evidence without running Agents

Assess saved trajectories with a model

agentbench exp3 judge auto

agentbench exp3 judge auto [-h] [--model MODEL] [--env-file ENV_FILE]
                                  --experiment EXPERIMENT [--root ROOT]

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`--model`Option	string	No	—	Judge model override on the configured provider
`--env-file`Option	Path	No	—	—
`--experiment`Option	Path	Yes	—	—
`--root`Option	Path	No	<checkout>	—

Terminal help: agentbench exp3 judge auto --help

Export evidence and a manual review template

agentbench exp3 judge export

agentbench exp3 judge export [-h] --output OUTPUT --experiment
                                    EXPERIMENT [--root ROOT]

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`--output`Option	Path	Yes	—	—
`--experiment`Option	Path	Yes	—	—
`--root`Option	Path	No	<checkout>	—

Terminal help: agentbench exp3 judge export --help

Validate and retain a completed manual review

agentbench exp3 judge import

agentbench exp3 judge import [-h] --input INPUT --experiment EXPERIMENT
                                    [--root ROOT]

Parameters
Parameter	Value	Required	Default	Description
`-h, --help`Option	flag	No	—	show this help message and exit
`--input`Option	Path	Yes	—	—
`--experiment`Option	Path	Yes	—	—
`--root`Option	Path	No	<checkout>	—

Terminal help: agentbench exp3 judge import --help

apiJSON command-line interface for scripts and external testing agents. Commands return structured JSON on stdout; execution logs go to stderr.

Read a registered agent’s source revision, framework, status, case limits, and task context.

agentbench api agent

agentbench api agent [-h] agent_id

Parameters
Parameter	Value	Required	Default	Description
`--root`OptionBefore subcommand: agentbench api	Path	No	<checkout>	—
`--env-file`OptionBefore subcommand: agentbench api	Path	No	—	—
`-h, --help`Option	flag	No	—	show this help message and exit
`agent_id`Positional	string	Yes	—	—

Terminal help: agentbench api agent --help

Prepare an agent’s source at a specified Git revision and return the prepared target metadata without running test cases.

agentbench api prepare

agentbench api prepare [-h] --revision REVISION agent_id

Parameters
Parameter	Value	Required	Default	Description
`--root`OptionBefore subcommand: agentbench api	Path	No	<checkout>	—
`--env-file`OptionBefore subcommand: agentbench api	Path	No	—	—
`-h, --help`Option	flag	No	—	show this help message and exit
`agent_id`Positional	string	Yes	—	—
`--revision`Option	string	Yes	—	—

Terminal help: agentbench api prepare --help

Run an abb.cases.v1 case collection against one agent, save execution evidence, and optionally invoke a Judge.

agentbench api execute

agentbench api execute [-h] --case-file CASE_FILE
                              [--results-dir RESULTS_DIR] [--judge JUDGE]
                              [--model MODEL] [--timeout TIMEOUT]
                              [--max-steps MAX_STEPS] [--revision REVISION]
                              agent_id

Parameters
Parameter	Value	Required	Default	Description
`--root`OptionBefore subcommand: agentbench api	Path	No	<checkout>	—
`--env-file`OptionBefore subcommand: agentbench api	Path	No	—	—
`-h, --help`Option	flag	No	—	show this help message and exit
`agent_id`Positional	string	Yes	—	—
`--case-file`Option	Path	Yes	—	—
`--results-dir`Option	Path	No	—	—
`--judge`Option	string	No	—	—
`--model`Option	string	No	—	—
`--timeout`Option	float	No	2400	—
`--max-steps`Option	int	No	—	—
`--revision`Option	string	No	—	Prepare and execute this exact historical commit

Terminal help: agentbench api execute --help

Read a saved Suite’s result snapshot and run metadata without executing the agent again.

agentbench api result

agentbench api result [-h] --suite SUITE

Parameters
Parameter	Value	Required	Default	Description
`--root`OptionBefore subcommand: agentbench api	Path	No	<checkout>	—
`--env-file`OptionBefore subcommand: agentbench api	Path	No	—	—
`-h, --help`Option	flag	No	—	show this help message and exit
`--suite`Option	Path	Yes	—	—

Terminal help: agentbench api result --help

Read one run’s validated Judge request, derived trajectory, trace spans, and evaluation data.

agentbench api evidence

agentbench api evidence [-h] --suite SUITE --run-id RUN_ID

Parameters
Parameter	Value	Required	Default	Description
`--root`OptionBefore subcommand: agentbench api	Path	No	<checkout>	—
`--env-file`OptionBefore subcommand: agentbench api	Path	No	—	—
`-h, --help`Option	flag	No	—	show this help message and exit
`--suite`Option	Path	Yes	—	—
`--run-id`Option	string	Yes	—	—

Terminal help: agentbench api evidence --help

Read the text content and metadata of a registered Judge evidence attachment from a saved run.

agentbench api attachment

agentbench api attachment [-h] --suite SUITE --run-id RUN_ID --path
                                 PATH

Parameters
Parameter	Value	Required	Default	Description
`--root`OptionBefore subcommand: agentbench api	Path	No	<checkout>	—
`--env-file`OptionBefore subcommand: agentbench api	Path	No	—	—
`-h, --help`Option	flag	No	—	show this help message and exit
`--suite`Option	Path	Yes	—	—
`--run-id`Option	string	Yes	—	—
`--path`Option	string	Yes	—	—

Terminal help: agentbench api attachment --help

Validate and save an external Judge report for a completed, host-accepted execution with no assigned Judge.

agentbench api submit-judge

agentbench api submit-judge [-h] --suite SUITE --run-id RUN_ID --report
                                   REPORT

Parameters
Parameter	Value	Required	Default	Description
`--root`OptionBefore subcommand: agentbench api	Path	No	<checkout>	—
`--env-file`OptionBefore subcommand: agentbench api	Path	No	—	—
`-h, --help`Option	flag	No	—	show this help message and exit
`--suite`Option	Path	Yes	—	—
`--run-id`Option	string	Yes	—	—
`--report`Option	Path	Yes	—	—

Terminal help: agentbench api submit-judge --help

Match saved Suite execution evidence against the agent’s ground-truth failures without rerunning the agent.

agentbench api gt

agentbench api gt [-h] --suite SUITE

Parameters
Parameter	Value	Required	Default	Description
`--root`OptionBefore subcommand: agentbench api	Path	No	<checkout>	—
`--env-file`OptionBefore subcommand: agentbench api	Path	No	—	—
`-h, --help`Option	flag	No	—	show this help message and exit
`--suite`Option	Path	Yes	—	—

Terminal help: agentbench api gt --help

Generated from the ABB CLI parser. Options inherited from a parent command must appear before its subcommand. A dash means no default or description is specified by the CLI.