Skip to content

ztd-cli Review Authority Model Draft

Draft authority model for ztd-cli concept, policy, generated artifacts, AI review, and CLI checks.
id ztd-cli-review-authority-modelformat draft markdown
draft
validation: draftcoverage: partialopen questions present
authority classes: presentgenerated view boundary: presentAI boundary: presentrules index: not registered

This document defines who owns review authority for the target ztd-cli direction. It is not the Concept Spec body and does not replace human approval.

Purpose

The ztd-cli direction depends on a clear split between human-owned intent, AI-assisted review, CLI-owned mechanical checks, and generated artifacts.

This model prevents generated output, AI summaries, or implementation convenience from becoming the source of truth for the package concept or policies.

Authority Classes

Human-owned Intent

Humans own durable package direction and acceptance of concept/policy changes.

Primary examples:

  • Package Concept.
  • Testing Policy.
  • Review Authority Model.
  • Technology Policy.
  • Issue goals and explicit human decisions.
  • Decisions to deprecate, split, or rename packages.

AI may draft, compare, ask questions, and propose wording. Those proposals are review input until accepted by a human.

AI-led Review Management

AI may lead review workflow and consistency checking.

Primary examples:

  • Finding contradictions between Package Concept, Testing Policy, Technology Policy, and current implementation.
  • Separating current-state facts from target-state concept statements.
  • Proposing staged migration plans.
  • Reporting risks, blockers, and open decisions.

AI review output is not policy. It must cite or point to the human-owned source that supports the claim.

CLI-owned Mechanical Checks

CLI checks own deterministic mechanical detection.

Primary examples:

  • Generated mapper drift checks.
  • Generated project smoke checks.
  • Structured metadata checks.
  • Generated review packets or reports.
  • Dependency and runtime-free verification scripts once they exist.

CLI output is evidence, not final judgment. When a generated view is wrong, the source artifact or generator should be fixed.

Generated Artifacts

Generated artifacts are machine-owned outputs.

Primary examples:

  • AOT row mappers.
  • Generated test plans.
  • Generated analysis files.
  • Generated review pages.
  • Generated DAO scaffolds when marked as generated.

Generated artifacts must remain reproducible from source artifacts. Humans review their meaning through source SQL, DDL, query contracts, policies, tests, and drift checks.

Required Review Posture

Reviews must distinguish:

  • target concept versus current implementation;
  • human-approved policy versus AI proposal;
  • generated evidence versus source of truth;
  • runtime-free standard path versus optional advanced runtime libraries.

Open Questions

  • Which ztd-cli policy documents should become structured metadata-backed pages?
  • Which CLI checks should be considered merge-blocking once the runtime-free path is implemented?
  • Who owns acceptance of advanced runtime package boundaries?

Released under the MIT License.