Reference

CLI

Every command the CLI supports, every flag that's wired to behavior. No dead parameters, the codebase enforces that every declared flag drives runtime logic.

Commands

ifixai run

Run the test suite (Standard or Full mode).

FlagPurpose

--model / -mModel identifier (defaults depend on provider)

--api-key / -kOverride env-var-sourced credentials

--endpoint / -eHTTP endpoint URL (when --provider http)

--modestandard (default) | full

--eval-modedeterministic | single | full | self (default: auto-detect cross-provider)

--fixtureName of a built-in fixture or path to a YAML file

--test / -bRun a single test by ID (B01…B32)

--strategicRun the top 8 strategic tests only

--min-scoreCI gate; exit non-zero when overall < threshold (default 0.85)

--judge-providerOne or more distinct judge providers (Full mode)

--judge-api-keyJudge credentials, arity must match --judge-provider

--judge-modelJudge model identifiers, arity must match --judge-provider

--judge-budgetMaximum judge calls per run (default 0 = unlimited)

--nameLogical name for the system under test

--output / -oReport directory for scorecards (default ./ifixai-results/)

--reliability-outWhere manifest.json and reliability.json are written (default runs)

--formatjson | markdown | both (default both)

--timeout / -tPer-call timeout in seconds (default 30)

--groundingHow the SUT receives governance context: sut | fixture | none (default sut)

--concurrency / -jParallel calls 1–20 (default 5; env IFIXAI_CONCURRENCY)

--no-parallelAlias for --concurrency 1

--dry-runEstimate inspection and judge call counts, then exit without making provider calls

--sut-temperatureTemperature for the system under test (default 0.0)

--sut-seedSampling seed for the system under test

--b12-seedSeed for B12 prompt-injection corpus shuffling (default 20260422)

--b14-seedSeed for B14 covert-side-task corpus shuffling (default 20260422)

--b30-seedSeed for B30 deployer-rule corpus shuffling (default 20260422)

ifixai init

Onboarding wizard: verifies the smoke fixture exists, detects which provider env vars are set, and prints a suggested first-run command.

ifixai list tests

List all 32 tests with ID, category, threshold, weight, mode.

ifixai list fixtures

List the available built-in fixtures.

ifixai validate <file>

Validate a fixture YAML/JSON file against the schema. Non-zero exit on failure.

ifixai compare <baseline> <enhanced>

Vendor-neutral comparison report between two scorecard.json files.

Common recipes

Local smoke test

bash

ifixai run --provider openai --model gpt-4o-mini --strategic

Single-test debugging

bash

ifixai run --provider openai --test B12 --fixture ./my-fixture.yaml

CI regression gate

bash

ifixai run --provider openai --strategic --min-score 0.85

Full-mode leaderboard run

bash

ifixai run --mode full \
  --provider openai --model gpt-4o \
  --fixture ./my-fixture.yaml \
  --judge-provider anthropic --judge-api-key $ANTHROPIC_KEY \
  --judge-provider gemini    --judge-api-key $GEMINI_KEY

Vendor-neutral comparison

bash

ifixai compare \
  runs/r-8c4f2e1d/scorecard.json \
  runs/r-9d5a3f2e/scorecard.json

Exit codes

0Success. All tests completed and the overall score met --min-score.

1Invalid arguments, connection failed, or mode conflict. Fix flags or env vars.

2Score below --min-score, or no scorable evidence. CI-gate-failure signal.

✓

Every flag is wired

A code-style rule in the project: no dead CLI parameters. Every flag listed here drives runtime behavior somewhere in the codebase. If you find one that doesn't, it's a bug.