The skill file lived in the main repo, which meant `npx skills add`
had to clone the entire RenderCV codebase and exposed internal dev
skills. A dedicated lightweight repo solves both problems.
- Create rendercv/rendercv-skill repo as a read-only distribution channel
- Add it as a submodule at .claude/skills/rendercv-skill/
- Update generation script and tests to write to the submodule path
- Add submodules: true to test workflow checkout steps
- Update docs, README, and install commands to use rendercv/rendercv-skill
- Add --recursive clone instruction to developer guide
- Delete the old skills/ directory at repo root
- Add SKILL.md for AI agents: YAML schema reference, entry type field
tables, design samples per theme, CLI reference, and important patterns
(YAML quoting, phone validation, bullet characters)
- Add Jinja2 template and generate.py that auto-generates SKILL.md and
docs/llms.txt from live Pydantic models and sample generators
- Add promptfoo eval suite (15 tests across 4 files): cv_generation,
pdf_parsing, design, and cli_workflow
- Add deterministic grader that validates LLM output through RenderCV's
own pydantic pipeline (not jsonschema)
