dsh-xingtu-skills

AI agent skill configurations are often scattered across tools, forcing developers to manually copy SKILL.md files every time they switch environments. dsh-xingtu-skills bundles 26 battle-tested agent skills into a single DeepSeek Harness plugin. Once installed, DSH loads the full skill catalog into the session automatically and triggers each skill on demand through progressive disclosure, eliminating manual copying and per-skill activation.

The skills span three practice areas: the Caveman series (13 skills) tackles token compression and context management to squeeze more signal out of every prompt; the Ponytail series (6 skills) enforces codebase minimalism and technical-debt awareness, nudging the agent to question necessity before building; and the engineering-practice series (7 skills) covers a complete workflow from investigation through safe refactoring to verified stop. Every skill follows the cross-tool name, description, and when_to_use convention and is also compatible with Claude Code, CodeBuddy, Codex, Cursor, and Gemini CLI.

The plugin ships with zero runtime dependencies and passes both static checks and a clean-environment boot test, so it integrates cleanly without pulling in extra packages. For solo developers and teams who want a single, consistent skill layer that improves token efficiency, code quality, and engineering discipline simultaneously, this is a one-command layer on top of their existing workflow.