dsh-skill-creator

In a DeepSeek Harness session, turning a conversation, a document, or a verbal process description into a reusable, distributable skill definition used to mean hand-writing frontmatter, repeated format alignment, manual packaging, and polishing trigger descriptions by feel. dsh-skill-creator compresses that pipeline into four tool calls: capture intent and generate a well-structured SKILL.md draft, validate the directory against six compliance checks, package a .skill artifact after validation passes, and rewrite the description to improve hit rate while preserving every original trigger term. The entire flow works offline with zero external runtime dependencies.

The core capabilities span the full skill lifecycle. skill_new extracts steps, tools, input/output formats, and boundary conditions from a goal string or the current session context, returning an intent summary and clarification questions in dry-run mode before writing the final SKILL.md. skill_validate reports PASS/FAIL for each of six rules. skill_package enforces validation before zipping into a store-compressed .skill file, automatically excluding pycache, node_modules, .DS_Store, and similar artifacts. skill_improve_description extends scenario vocabulary while guaranteeing that every original trigger word survives in the rewritten description; in commit mode only the description line in frontmatter is replaced, leaving the rest byte-identical.

Ideal for developers building or maintaining DSH skill libraries, teams that need to quickly codify repeated operations into distributable skills from within a session, and solo builders who want to cut down on manual YAML checking and packaging errors. The plugin self-registers a runtime provider at apply time, so generated skills are immediately listable and loadable by the skill tool without any profile configuration changes.