skill-authoring

What problem does it solve? Writing an Agent Skill that actually triggers, loads, and improves agent behavior is hard: descriptions fail to route, bodies exceed context budgets, references never get read, and rules restate what the model already knows. This Skill provides the specification constraints, authoring rules, evaluation methodology, and publishing mechanics to build skills that measurably close real baseline gaps. ## Core Features & Use Cases - Specification-grounded authoring: Enforces the agentskills.io frontmatter contract (name rules, description limits, allowed fields) plus body budgets, one-level reference depth, and progressive disclosure tiers. - Evaluation-driven development: Requires at least three scenarios with a near-miss negative and a measured no-skill baseline before writing body text, with grading schemas and variance guidance. - Review and optimization workflows: Structured checklists for reviewing existing skills, optimizing descriptions with held-out trigger query sets, and a two-agent iteration loop that reads navigation signals. - Publishing and provenance: Covers installer scan rules, flat repository layout, per-directory licensing, SOURCES.yaml provenance records, and generated NOTICE attribution. - Use Case: A team repeatedly re-explains internal GraphQL gateway rules to their agent; use this Skill to baseline the failures, draft a minimal skill closing only those gaps, and verify with trigger-rate measurements before publishing it to the team repository. ## Quick Start Ask the agent to review your existing SKILL.md file for spec compliance, description quality, and reference structure, and to report actionable findings with replacement text.