Entity Creator Skills Overhaul: Orchestrator, Template References, Codegen Gaps

The entity-creator skills ported to v2 in 2026-05-21-entity-creator-skills-v2-migration.org have three remaining problems:

1. No orchestrator. Asking an LLM to "add a new entity" invokes at most one layer skill. There is nothing that says "a complete entity requires all seven layers, and you must execute all of them." The result is entities with missing layers.
2. Pattern docs contain code snippets. The knowledge docs created in the v2 migration embed #+begin_example C++ blocks. These snippets drift from the codebase, and they tempt the LLM to hand-write code instead of using codegen. The mustache templates in projects/ores.codegen/library/templates/ are the ground truth; the docs should reference them, not duplicate them.
3. Codegen gaps are invisible. Four layers — HTTP, CLI, Shell, Wt — have no codegen profiles and no mustache templates. The layer skills treat this as normal and describe a manual workflow. The correct response is: treat the absence of a profile as a bug, document the gap explicitly in the skill, and require the LLM to implement the profile before generating.

Additionally there is no documentation covering Qt menu and toolbar wiring (plugin on_login() registration, MDI window toolbar construction), which is a required step when adding the Qt layer for any entity.

Codegen gaps (no profile, no templates):

Create Entity Creator Skill using the codegen script.

The skill describes a complete regular entity as requiring all seven layers, executed in dependency order. Each layer is a separate task the LLM creates with TaskCreate and resolves by invoking the appropriate sub-skill.

Add two new sections:

1. Complete entity file checklist: the canonical list of files a regular entity must have across all layers (modelled on the currency entity). This is what reviewers check when a "complete entity" story is marked done.
2. Template references: for each layer that has a codegen profile, list the mustache templates by name (e.g. cpp_domain_type_class.hpp.mustache). The templates are the ground truth; the org doc points to them, it does not reproduce them.

For each of the seven layer knowledge docs created in the v2 migration:

For docs with templates: replace #+begin_example code blocks with a table of template names and their output paths (taken from profiles.json). Add a note: "Read the template directly for the authoritative implementation — do not reproduce it here."

For gap docs: replace the layout pseudo-code with:

⚠ *Codegen gap*: no profile exists for this layer.  Before creating
any {layer} entities:

1. Create mustache templates in [[proj:projects/ores.codegen/library/templates][=projects/ores.codegen/library/templates/=]].
2. Add a ={layer}= profile entry to [[proj:projects/ores.codegen/library/profiles.json][=projects/ores.codegen/library/profiles.json=]].
3. Validate against an existing entity (e.g. =currency=).
4. Only then use codegen to generate the layer.

Until the profile exists, see the recipes for the manual path — but
treat this as a workaround, not the target state.

Update all seven layer skills to:

1. Add a Templates section listing the mustache templates the profile runs (for codegen-covered layers).
2. Add a Codegen gap section (for HTTP, CLI, Shell, Wt) that names the gap and states the protocol (stop → raise story → implement profile → then generate).
3. Remove any remaining inline code from the skill itself (skills only describe when and how; the templates carry the what).

• Push feature/doc-version-marker, open the PR.
• PR title: Entity creator: orchestrator, template refs, codegen gap protocol
• No "Test plan" section.

After every commit:

1. ./projects/ores.codegen/validate_docs.sh — must pass.
2. No #+begin_example or #+begin_src blocks remain in any skill file.
3. No #+begin_example or #+begin_src blocks remain in any layer knowledge doc.
4. Every layer skill has either a Templates or Codegen gap section.
5. The orchestrator skill table has exactly seven layer rows.