maida run#

Runs a traced Python agent repeatedly in fresh copies of the current tracked and nonignored workspace, evaluates every resulting trace with the selected policy, and aggregates the outcomes into PASS, FAIL, or INCONCLUSIVE.

maida run AGENT_SCRIPT [options]

Argument/Option

Default

Description

AGENT_SCRIPT

required

Traced Python script inside the current Git workspace

--trials

policy value (3)

Number of isolated subprocess trials

--confidence-level

policy value (0.95)

One-sided Wilson coverage (z = 1.645 at 0.95)

--pass-rate-threshold

policy value (0.90)

Required pass rate

--baseline, -b

-

Baseline JSON applied to every trial

--policy

.maida/policy.yaml

Assertion and statistical gate settings

--format, -f

text

text, json, or verdict-first markdown

--fail-fast / --no-fail-fast

policy value (true)

Stop on an irreversible blocking failure, or force the full fixed-N sample

--json-out

-

Atomically write report schema 2.0.1 to a sidecar

maida run my_agent.py --baseline .maida/baselines/my_agent.json \
  --policy .maida/policy.yaml --format markdown --json-out maida-report.json

Each trial must create exactly one completed trace. Exit 1 is reserved for FAIL; PASS and the provider-neutral INCONCLUSIVE verdict exit 0, so CI consumers must read the JSON verdict rather than infer uncertainty from the process status. Missing inputs exit 2 and internal execution failures exit 10.