dsh-skill-adapter

The Claude Code and Codex ecosystems have already accumulated tens of thousands of SKILL.md skills whose frontmatter follows the Anthropic convention (whenToUse, disable-model-invocation, user-invocable, and more). DeepSeek Harness natively understands /SKILL.md files and its key names align with that same spec, but the middle layer of import, normalization, and distribution simply did not exist. You could not practically pull those community skill libraries into DSH. dsh-skill-adapter closes that gap with a zero-dependency YAML-subset parser that silently reconciles key-name differences, converts invalid names to kebab-case, and emits a clear compatibility report whenever required fields are missing.

The toolchain offers two integration paths. A CLI batch importer recursively fetches skills from a GitHub repository or a local directory, validates and normalizes each one, then writes them into DSH's skills directory so the agent discovers them automatically on next launch. Alternatively, a DSH plugin registers a SkillProvider at runtime via ctx.skills.registerProvider, mounting skill sources dynamically without any file copy. This is ideal when your skill set evolves frequently and you want zero-downtime refreshes. Both paths share the same core parsing and normalization logic, guaranteeing consistent behavior.

If you have been building a personal skill library in Claude Code or Codex, or you want to reuse high-quality community skill collections without hand-editing every frontmatter key, dsh-skill-adapter lets you migrate an entire ecosystem into DeepSeek Harness with a couple of configuration lines. Spend your time designing skills, not plumbing formats.