Constitutional skill must own doctrine while execution skill stays subordinate

Problem

When a repo needs both architecture doctrine and practical implementation guidance, it is tempting to put both into one authoring skill.

That fails in two common ways:

• the execution skill becomes a kitchen sink
• a second doctrine skill gets created but drifts because nothing actually routes work through it

Symptoms

• The “practical” skill starts containing architecture law, public API shape, precedence rules, and anti-pattern catalogs.
• The “doctrine” skill reads well but lacks routing, ownership, and update enforcement, so people ignore it.
• Review lanes cannot answer simple binary questions like:
  • who owns this decision?
  • should this route upward first?
  • what counts as doctrine reaffirmation?

What Didn't Work

• Keeping doctrine and execution mechanics in one skill.
• Splitting the skills without a hard routing rule.
• Relying on readers to “just know” when to consult the doctrine skill.
• Verifying only section presence instead of behavior-level routing and ownership.

Solution

Use a constitutional source-of-truth skill for reusable API and architecture decisions, and keep the execution skill explicitly subordinate.

The constitutional skill should own:

• reusable architecture doctrine
• public API shape decisions
• runtime/service-boundary patterns
• ownership/layering law
• performance/scalability law
• anti-patterns

The execution skill should own:

• implementation mechanics
• file placement
• wrapper patterns
• typing details
• execution flow

Then make the split enforceable:

• ambiguous cases route upward to the constitutional skill first
• reusable API/refactor lanes must include either:
  • updated
  • or reaffirmed: <section>
• the execution skill must not restate long-form doctrine
• a routing matrix and ownership table must exist in the constitutional skill

Why This Works

The split becomes governable instead of aspirational.

• The doctrine stays in one place.
• The execution companion stays usable.
• Reviewers can reject drift with binary checks.
• Future architecture work has a clear place to update or reaffirm.

Prevention

• If a new skill is meant to own doctrine, make it the constitutional source of truth explicitly.
• If a companion skill exists, add a hard routing gate and a short owner map.
• Do not let the execution companion restate long-form doctrine “for convenience.”
• Verify the split with routing scenarios, ownership checks, and explicit reaffirmation evidence, not just section existence.