dsh-skill-authoring

DSH skill mechanics hide several counter-intuitive traps: one stray space in frontmatter and the skill silently vanishes; the rank 100-to-600 scan order decides which same-named skill wins; and description is the only field that ever reaches the model, so its wording directly determines whether a trigger fires. dsh-skill-authoring bundles those scattered rules into a single installable skill, giving your agent a reference manual for writing, reviewing, and refactoring skills instead of relying on trial and error.

The skill covers the full spectrum from static structure to runtime behavior. On the static side it documents frontmatter required and optional fields with their legal boolean syntax, skill-root priority and the scope of change detection, how to stack trigger cues and write fallback boundaries in description, the 300-to-500-line target range, and when to split into references/. A zero-dependency Node validation script runs six checks in one pass-frontmatter, required fields, file references, in-page anchors, cross-references, and length-supporting JSON output for CI pipelines. On the behavioral side it prescribes an A/B method: do the same task with and without the skill, compare the actual artifacts rather than self-reports, and codify the discipline you must hard-code when delegating to child agents.

This plugin is aimed at developers writing or maintaining DSH skills, and at teams that want a systematic understanding of agent skill loading to reduce silent failures and missed triggers.