Sub-skill: Turn Your Messy Skill Library Into a Hierarchy

Where this fits

You've been adding skills for weeks. Now your agent's skills/ folder looks like a thrift store — pdf-extractor, pdf-summarizer, pdf-visual-inspector all sitting side by side, no structure. Every time you prompt for a PDF task, the model has to scan dozens of loosely-related skills, hoping it picks the right one. You need a way to group them by domain, but manually moving folders and rewriting frontmatter is error-prone and tedious. This is exactly where sub-skill steps in: it reviews your inventory, proposes parent-child groupings, and then (after your explicit approval) reorganizes the skills into has-sub-skill: true bundles with backups.

Why agents benefit

• Proposes meaningful groupings – sub-skill.review reads your skill metadata and suggests bundles based on descriptions, names, and usage patterns—not random clustering.
• Requires human approval – Consolidation only runs after you confirm the plan, so you never lose control over your skill structure.
• Creates timestamped backups – Before moving anything, it saves your current state, making rollback trivial if something breaks.
• Handles the mechanics – It updates frontmatter (has-sub-skill: true), moves files, and can even nest existing sub-skills under new parents.
• Keeps the inventory discoverable – By grouping related skills under one parent, the model's next search space narrows, improving selection accuracy on subsequent runs.

Practical scenarios

Cleaning up a growing PDF toolkit

You have 12 skills for parsing, splitting, merging, and extracting text from PDFs. sub-skill.review suggests a parent pdf-toolkit with each PDF skill as a child. After approval, sub-skill.consolidate moves them into pdf-toolkit/ and sets has-sub-skill: true on the parent. Now a single pdf-toolkit invocation can delegate to the right child.

Merging data-loading skills after a hackathon

You and a teammate built separate skills for CSV, JSON, and Parquet loading during a sprint. They overlap. The review flag suggests a data-loaders parent, and you consolidate—keeping your own version, but with a clear namespace.

Evaluating a new skill before adding it

You're about to add summarize-slides from a community repo. sub-skill.review can tell you if it fits under an existing presentation-tools parent or should stay standalone. This prevents future clutter before it happens.

Add it to your agent workflow

Install with:

npx skills add wangyuahn/nori-code --skill sub-skill

Then trigger a review:

User: Review my skill inventory and propose groupings.
Agent: Runs sub-skill.review → outputs a tree of suggestions.
User: Approve the `pdf-toolkit` grouping.
Agent: Runs sub-skill.consolidate → moves files, creates backup.

If you're curating a set of skills for your team or want to see how this skill compares to others, check out its entry on OpenAgentSkill.

Compare before adopting

Before you add sub-skill to your permanent stack, compare it against:

• Manual folder organization – If your inventory is under 10 skills, a script might be overkill.
• Other skill-management tools – Look for alternatives that also support backup, dry-run, and explicit approval steps.
• Maintenance freshness – Check the repo's last commit and issue responses; this project has only 13 stars, so responsiveness matters.
• Workflow fit – Does it align with how you already name and structure skills? If your habits differ, you may need to adapt the proposed groupings.

Why it is worth tracking

Quality signals are modest but promising: the skill implements a clean two-phase pattern (review/consolidate) and respects user confirmation—a solid approach to agent-driven file changes. The low star count suggests early development, but the design is simple enough to fork and customize. Track it if you're accumulating skills beyond a handful, or if you want a template for building container skills. Evaluate it again when the repo gains more stars or adds support for nested inventories beyond one level.