Summary

Proposal to add two optional frontmatter fields that declare relationships between skills:

• prerequisite-skills: Skills that should ideally run before this skill
• related-skills: Complementary skills that work well together

Motivation

Skills often work best in combination. For example, a blog-generator skill produces better output when:

1. An audience-builder skill has first defined the target audience
2. A brand-voice-generator skill has established tone guidelines

Currently, there's no standard way to express these relationships. Agents must either:

• Hardcode knowledge about skill combinations
• Rely on users to manually sequence skill execution
• Hope the LLM figures it out from context

Declaring relationships in frontmatter enables agents to:

• Suggest or auto-load prerequisite skills
• Present related skills as follow-up options
• Make informed decisions about execution order

Proposed Format

---
name: blog-generator
description: Creates blog posts optimized for your audience and brand voice
prerequisite-skills:
  - slug: audience-builder
    reason: "Defines target audience characteristics for tailored content"
  - slug: brand-voice-generator
    reason: "Establishes tone and style guidelines"
related-skills:
  - slug: seo-optimizer
    reason: "Improves search visibility after content is created"
  - slug: image-generator
---

Field Definitions

Each entry contains:

• slug (required): The skill's name/slug identifier
• reason (optional): Brief explanation of the relationship (helps agents reason about when/why to use)

Simple Format Alternative

For cases where reasons aren't needed:
prerequisite-skills:
- audience-builder
- brand-voice-generator

Design Considerations

1. Hints, not requirements: These are suggestions for agents, not hard dependencies. Skills should still function independently.
2. Reason field for LLM reasoning: The optional reason helps LLMs understand why skills are related, improving their decision-making.
3. No circular dependency validation: Keep it simple - let implementations decide how to handle cycles.
4. Forward references allowed: Skills can reference other skills that may not exist yet (useful during development).

Prior Art

• npm/package.json: dependencies, peerDependencies, optionalDependencies
• Python/pyproject.toml: dependencies, optional-dependencies
• MCP: Server composition patterns

Implementation

We've implemented this in our no-code agent building platform Agentman.ai and also made it available in Claude, Claude Code and other MCP client compatible platforms through myAgentSkills.ai. We found it significantly improves multi-skill workflows. Happy to share learnings.