This catalog has a public incubating release. Skills can be used directly from the checkout or copied into an agent-specific skills location.
For reproducible use, pin a tag or commit SHA. The initial incubating catalog release is tagged as v0.1.0-incubating.
Use the repository instructions from AGENTS.md. For local skill testing, copy or symlink individual skill folders into your Codex skills directory, then invoke them by name, for example:
Use $uxl-onemath to diagnose this oneMath CMake failure.
Run validation first:
python scripts/validate_catalog.py
python scripts/run_evals.py --validateClaude Code reads CLAUDE.md; this file points to AGENTS.md. Start Claude Code from the repository root so it loads the repository instructions.
Use explicit prompts while skills are incubating:
Use the uxl-onemath skill at skills/uxl-onemath to choose a dispatch model for this code.
Copilot reads .github/copilot-instructions.md and path-specific instructions in .github/instructions/.
For cloud-agent work, include the repository and ask Copilot to inspect:
AGENTS.mdskills.yaml- The relevant
skills/<skill-name>/SKILL.md - The relevant skill card
The repository includes thin wrapper files for these tools. Keep AGENTS.md canonical and regenerate wrappers after changing repository-wide guidance:
python scripts/generate_agent_wrappers.pyThe repository includes .agents/skills/uxl-skills-catalog.md as a lightweight repository skill wrapper. The primary progressive-disclosure skills remain under skills/<skill-name>/.
- Keep incubating work in this catalog.
- Get project maintainer review for each library-specific skill.
- Move the reviewed skill source of truth into the owning project repository.
- Mirror the reviewed skill back into this catalog.
- Publish tags and ask users to pin by tag or commit SHA.
See docs/release-and-promotion-policy.md before promoting any skill beyond incubating or pilot.