AgentSkillsCN

create-skill

以确定性的架构、工作流程、安全规范与质量检查为基础，打造标准化的SKILL.md文档。

SKILL.md
--- frontmatter
name: "create-skill"
description: "Create canonical SKILL.md artifacts with deterministic schema, workflow, safety, and quality checks."
metadata:
  source_prompt: "prompts/create-skill.prompt.md"
  workflow_type: "generation"

Create Skill

Purpose

Generate complete SKILL.md artifacts for one skill folder, with explicit constraints and reusable references.

When To Use

  • •Use this skill when creating a new skill folder under skills/.
  • •Use this skill when regenerating a SKILL.md file to align with repository skill standards.
  • •Do not use this skill for validating an existing skill.

Required Inputs

  • •Skill slug (name) and target folder path.
  • •Capability summary and usage triggers.
  • •Required inputs, workflow outputs, and failure boundaries.
  • •Safety and portability constraints.

Workflow

  1. •Run discovery sequence in references/discovery-phases.md.
  2. •Apply schema and field rules from references/field-constraints.md.
  3. •Assemble required body sections using references/output-schema.md.
  4. •Apply safety constraints from references/safety.md.
  5. •Produce final artifact using references/output-format.md.
  6. •Validate against references/quality-checklist.md before final output.

Output Expectations

  • •One complete SKILL.md artifact at skills/{name}/SKILL.md.
  • •Frontmatter and folder-name parity are correct.
  • •Body sections are complete, actionable, and concise.
  • •Output remains provider-neutral and portable.

Resources

  • •Discovery workflow: references/discovery-phases.md
  • •Field constraints: references/field-constraints.md
  • •Output schema: references/output-schema.md
  • •Safety policy: references/safety.md
  • •Output contract: references/output-format.md
  • •Quality checklist: references/quality-checklist.md
  • •Skill template: ../../templates/skill_template/SKILL.md

Constraints And Safety

  • •Keep skill instructions concise and execution-oriented.
  • •Avoid provider-specific lock-in wording.
  • •Use relative paths for local references.
  • •Treat allowed-tools as optional/experimental unless explicitly required.