AI agent skill advertises only 6 of 9 built-in themes

Problem

RenderCV 2.8 accepts nine built-in themes, but the generated AI agent skill advertises only six. The missing themes are ember, ink, and opal.

The generator intentionally limits detailed theme overrides to keep the skill context compact. However, the same hard-coded subset is also passed to the template as the complete available-themes list, so the skill incorrectly tells agents that only six themes exist.

Root cause

scripts/rendercv_skill/generate.py uses SKILL_THEMES both for:

1. the public list of available themes; and
2. the subset whose detailed override YAML is embedded.

These are separate concerns. The engine already exposes the authoritative dynamic list through available_themes.

Proposed fix

• Use rendercv.schema.models.design.built_in_design.available_themes for the skill metadata and Quick Start list.
• Rename the compact hard-coded subset to THEMES_WITH_DETAILED_OVERRIDES and use it only for expanded override examples.
• Clarify in the template that detailed examples cover a representative subset.
• Add a regression test comparing the themes advertised by the generated skill with the themes accepted by RenderCV.

I have implemented and tested this locally. The generated skill grows by only about 108 bytes because the three themes are listed without embedding their detailed override YAML.

Contribution workflow question

The generated SKILL.md lives in the read-only rendercv/rendercv-skill submodule, and the main repository's generated-file test requires that submodule content to be updated. External contributors do not have push access to the distribution repository, and it explicitly asks contributors to work in rendercv/rendercv.

What is the preferred workflow for a PR that changes the generated skill? I can open the main-repository PR immediately once there is a maintainer-supported way to publish the matching submodule commit.