cogniflow

cogniflow is the public launcher for Cogniflow installation and instance lifecycle operations. A real pip installation creates cogniflow.exe on Windows. Invoking it without arguments starts the installed sibling Ratatui frontend; explicit commands use the scriptable Python command surface.

Local Repository Mode

The interactive frontend asks for exactly source checkout, target root, instance ID, and profile. It derives working artifact locations and forwards the equivalent explicit command:

cogniflow install --instance-id local-dev \
  --installation-root /opt/cogniflow-instance \
  --cogniflow-home /home/user/.cogniflow \
  --mode local-repository --checkout-path /work/cogniflow \
  --artifact-output-root /tmp/cogniflow-artifacts \
  --external-wheelhouse-path /tmp/cogniflow-external \
  --profile urn:cf:profile:minimal

The launcher discovers an installed orchestrator from cogniflow.install_orchestrators.v1, verifies its matching console script, and crosses a versioned JSON-lines process boundary. It coordinates the existing lifecycle owners but does not implement profile selection, artifact resolution, target transactions, managed-tool provisioning, or semantic service policy itself.

PyPI Mode accepts only the bundled exact release profile, cogniflow at the matching release version, and https://pypi.org/simple. The public installation is python -m pip install cogniflow; local wheelhouses, private indexes, editable metadata, and source checkout injection are not accepted as public-release proof. Publication and isolated release-validation procedures are documented in ../docs/scb-17-public-release.md.

Lifecycle commands

cogniflow repair exposes inspect, preview, and apply through semantic lifecycle services. cogniflow update previews or applies profile transitions. cogniflow uninstall previews and applies authenticated instance teardown. Provider and orchestrator listing inspect distribution metadata without executing providers.

Exit codes are 0 for success, 2 for CLI errors, 3 for discovery errors, 4 for process/protocol errors, and 5 for structured provider errors. Bootstrap discovery uses installed metadata and local semantic documents because the production runtime graph and Fuseki may not yet exist.