Check an operation against explicit completion obligations. The kernel records which artifact checks passed, failed or still require human judgment, and includes durable verification/resume records for longer filesystem workflows.
Python 3.11 or newer. The package depends on Pydantic 2.
python -m pip install -e .
python -m examples.demoexamples/demo.py defines one generic report rule without using an archive-specific profile. A missing report fails, a TODO stub fails, and a substantive synthetic report passes. Changing the check to human review produces requires_review. Expected stages: ["fail", "fail", "pass"]; a JSON file preserves all three check/evidence snapshots.
Completion can be a record of specific satisfied and unresolved obligations rather than a free-form claim that the task is done. Read the mechanism and implementation notes for the specific boundaries and source links.
For persisted CLI verification, failure/resume records and a pending human check,
run python -m examples.cli_walkthrough and read the completion-gate guide.
A file-not-stub check does not prove factual accuracy. The CLI currently exits zero for both verified and partial; automation must inspect the recorded overall status and review_required field rather than equating exit zero with full verification. No model-judgment adapter is bundled.
python -m pytest runs the behavior tests (install pytest first). The runnable example above provides a separate first-use check.
MIT licensed; see LICENSE.md. Origin and release boundaries are documented in ORIGIN.md and SECURITY.md.
Open the saved synthetic result alongside its input and demonstration. The result is from the bundled synthetic example; local machine paths and temporary run identifiers are excluded from public projections.