Skills Overview: The Compound Edge

hub

Skills Overview: The Compound Edge

Every skill type explained. Built-ins, custom, plugin-injected. When to use a skill vs. a slash command vs. a hook.

The compound edge

A skill is a markdown file. The frontmatter declares when it activates; the body tells Claude exactly what to do. That's it. But the simplicity is deceptive: skills are where individual effort compounds into institutional memory.

Every time you encode a workflow into a skill file, you take it off your plate forever. The next time that situation arises (a new prospect, a weekly review, a client onboarding), you type a phrase and Claude does the work. The skill stack grows. The cognitive load shrinks.

Three skill types

Built-in skills are bundled in the vercel-plugin (or ManasFlow's skills pack) and injected automatically when their trigger phrases appear in the conversation. You never write them; they just appear.

Custom skills live in ~/.claude/skills/ or .claude/skills/ as plain .md files. The frontmatter declares triggers; the body is the instruction set. They load into every session and activate on matching phrases.

Plugin-injected skills are delivered dynamically through Claude Code's hook system. The plugin reads the project context (tech stack, open files, error logs) and pushes relevant skill content before the first token is generated.

Anatomy of a skill file

Every skill is a .md file with a YAML frontmatter block. The body is free-form instructions; write it the same way you'd brief a smart contractor.

Skill vs slash command vs hook

The three primitives are complementary, not interchangeable. The question is never "which one is better"; it's "which one fits the job."

Rule of thumb: If you'd put it in a runbook, make it a skill. If you'd put it in a Makefile, make it a hook. If you just want a shortcut label, make it a slash command.

Three worked examples

Real skills from the ManasFlow pack. Click any to see the frontmatter.

Distribution patterns

Skills need to reach the right sessions. There are four distribution patterns; pick the one that matches your scope.

Available in every project on your machine. Good for personal productivity skills.

Checked into git. Every teammate and CI run gets the same skill set.

Skills pushed at session start based on tech stack detection. Zero config for end users.

Reference skill files from CLAUDE.md with @path/to/skill.md. Keeps context lean.

The compound effect in practice: A team of 5 sharing a .claude/skills/ directory means every skill one person writes is available to everyone from the next git pull. 20 hours of workflow encoding becomes 100 hours of recovered capacity.

Want this running in your stack?

ManasFlow builds AI automation systems for production teams.

ruvnet Ecosystem: A Field Guide

Use Cases Gallery: 40+ Patterns