merceralex397-collab/meta-skill-engineering - 第 16 页

skill-evaluation

Evaluate whether a single skill routes correctly and produces better output than a no-skill baseline — measuring trigger precision/recall and output win rate. Use when someone says "is this skill working?", "validate before promoting", "does this skill still…

原文语言：英语

skill-lifecycle-management

Promote, deprecate, and track skills through lifecycle states (draft → beta → stable → deprecated → archived). Use when auditing maturity across a skill library, promoting a tested skill to stable, retiring a superseded skill, checking which skills are…

原文语言：英语

skill-provenance

Audit and record origin, authorship, evidence basis, license, and trust level for a skill. Use when someone asks "where did this skill come from?", "audit this skill's origin", "document provenance", or when evaluating external skills for adoption or…

原文语言：英语

skill-safety-review

Audit a SKILL.md and its bundled scripts for safety hazards — destructive operations missing confirmation gates, excessive permissions, prompt injection vectors, scope creep, and description-behavior mismatches. Use when a user says "review this skill for…

原文语言：英语

skill-trigger-optimization

Fix skill routing when the wrong skill fires or a skill won't fire at all. Use when the user says "this skill never fires", "wrong skill fired", "fix the triggers", "why isn't this skill being used?", or when a skill description is vague marketing copy that…

原文语言：英语

skill-variant-splitting

Split a broad skill into focused variants along stack, platform, scope, or domain axes. Use when "this skill does too much", "split this skill", "create variants for X and Y", a skill has disjoint "For X" / "For Y" sections, triggers on unrelated inputs, or…

原文语言：英语