Command line

Installing mlbundle adds an mlbundle console script with two commands. Both use the same readers as the Python API.

mlbundle inspect trained.mlbundle
mlbundle inspect trained.mlbundle --json
mlbundle verify trained.mlbundle

inspect

Validates metadata and integrity without importing any model library, so it works with the base install. The first line says what was and was not checked:

Metadata and integrity validated; model claims NOT verified.
Format: mlbundle 1
Bundle id: aa6f6b39-574a-499b-a0ff-de4cdd334050
Created: 2026-10-19T11:17:51.949658+00:00
Writer: "0.2.1"
Model: sklearn "DecisionTreeClassifier" (classifier)
Counts: features=2, outputs=1, classes=[2]
Features (declared): ["temperature", "humidity"]
Class labels (output-major): [["cold", "warm"]]
Target names (output-major): null
Members:
  manifest.json: validated (no self-integrity record)
  model.skops: 13065 bytes sha256=24d7cbab1d775ab279dc8ebc3c762b551b397ebf9b4917ac3c4474ff611b8813
  schema.json: 361 bytes sha256=ea62a6a7502ae8ea30b45b61bb18d8e31bd2ec55f026d9fbf0c05ef42d01ff25
  structure.json: 526 bytes sha256=66204da26781a4e2a800a1f3ed246ab1a43e8c97ced73d06a2640f897eeda03c
Training: {"python": "3.14.7", "implementation": "CPython", "os": "Windows", "os_version": "11", "architecture": "AMD64"}
Packages: {"catboost": "1.2.10", "mlbundle": "0.2.1", "mlschema": "0.3.0", ...}
ONNX environment: null
Present: schema=yes, structure=yes, samples=no
User metadata: {}

Composite pipelines add their step names, final estimator and transformed-feature boundary. Values in the human summary escape terminal control characters and are truncated at 512 characters, so a hostile bundle cannot rewrite your terminal.

--json prints the full validated manifest instead: lossless, ASCII-escaped JSON, suitable for jq or archiving.

verify

Loads the bundle through its adapter and recomputes every model-derived claim and the structure. It needs the model’s extra:

Valid bundle: sklearn "DecisionTreeClassifier"; verified format, integrity, schema/samples and native model claims/structure.

Errors and exit codes

Errors are a single bounded line on stderr naming the failure category. They never print bundle content or a traceback.

Exit code	Meaning	Example stderr
0	Success.
1	The bundle was rejected, or a dependency is missing.	mlbundle: InvalidContainerError: bundle rejected.
1		mlbundle: missing optional dependency; install mlbundle[xgboost].
2	Usage or I/O failure.	mlbundle: FileNotFoundError: I/O failed.
2		mlbundle: invalid arguments; use --help.

Use verify in CI or before publishing a bundle elsewhere; use inspect where model libraries are not installed or should not run.