Skip to main content

Architecture

qbit-ai-toolkit is the canonical, versioned asset source for Qbit consumers. It is not the Qbit CLI, not the Qbit Console, and not an application runtime.

qbit-cli ─┐
qbit-console ─┼── consumes versioned qbit-ai-toolkit assets
future consumers ─┘

Asset ownership

Assets are organized by type rather than by consuming product. A prompt, Docker template, installer, policy, skill, or script remains one canonical asset and declares its supported consumers in metadata.

Do not create parallel cli/, console/, or shared/ copies of the same logical asset. Consumer-specific behavior belongs in metadata, explicit variants, or a documented compatibility boundary.

Repository boundaries

The repository currently has two different AI-tooling concerns:

  1. Root .ai/, .agents/, .codex/, and .serena/ files are project-local tooling used to develop qbit-ai-toolkit itself.
  2. installers/codex-ai-tooling/ is a reusable versioned asset that emits equivalent tooling into consumer repositories.

These boundaries must not become implicit build dependencies. The project-local runtime must not self-install the reusable installer, and generated installer output must not replace canonical source.

Authority

When sources disagree, authority descends from:

  1. canonical versioned source and assets;
  2. schemas and catalog contracts;
  3. automated tests and installer state contracts;
  4. committed architecture and compatibility documentation;
  5. generated reports, graphs, caches, logs, and summaries.

Derived evidence may support analysis but does not become a source of truth.

Compatibility namespaces

The repository name is qbit-ai-toolkit. Existing installer state paths and managed-block markers containing qbit-toolkit are retained as versioned compatibility identifiers for the installer 1.0 contract. Renaming those identifiers requires an explicit migration and cannot be inferred from the repository rename.