| name | skill-builder |
| version | 1.0.0 |
| description | Build a new Agent Skill that follows the agentskills.io specification and best practices
— slim `SKILL.md` (≤ 200 lines / 5K tokens), valid kebab-case `name`, imperative
`description` under 1024 chars, progressive disclosure via `references/`, bundled
`assets/` and `scripts/`, and an MIT `LICENSE`. Use this skill whenever the user asks to
create, scaffold, build, write, or author a new Agent Skill — including phrasings like
"build a skill for X", "scaffold a new skill", "create an agent skill", "make me a skill
that does X", "write a SKILL.md for ...", or "I want to publish a skill on
agentskills.io". Also use when reviewing or refactoring an existing oversized `SKILL.md`
(a sign that detail should be moved to `references/`).
|
| license | MIT |
| compatibility | claude-code opencode |
| allowed-tools | ["Read","Write","Edit","Grep","Glob","Bash","AskUserQuestion"] |
| metadata | {"author":"MKAbuMattar","requirements":"Requires bash 4.0+ and `python3` (used by the validator for line/token counts). Generated skills target the open Agent Skills format (https://agentskills.io/specification)."} |
Skill Builder
Create new Agent Skills that follow the agentskills.io spec and the official best practices.
When to use
- The user asks to create, scaffold, build, or write a new Agent Skill.
- The user wants to publish a skill or refactor a bloated
SKILL.md.
- A task chain ends in "and put it in a skill so I can reuse it across projects".
- The auto-suggest hook (
scripts/task-tracker.py) fires a system-reminder noting that a similar task has been asked 3+ times — propose using this skill to package it.
- Self-monitor mode (always-on, even without the hook): track requests within the active conversation. When the user asks 3 substantively-similar tasks in a single session — the same kind of artifact, the same domain, the same workflow — proactively offer to package the recurring pattern as a skill. See Self-monitor mode below for the exact heuristic.
Required structure
Every skill is a folder with a SKILL.md. Optional directories carry detail loaded on demand:
<skill-name>/
├── SKILL.md # required: frontmatter + slim instructions (≤ 200 lines / 5K tokens)
├── LICENSE # MIT to match this repo
├── references/ # detailed docs the agent loads only when needed
├── assets/
│ ├── templates/ # - copy-edit starting points
│ └── examples/ # - full reference implementations
└── scripts/ # bundled executables (validators, scaffolders, processors)
Frontmatter must include name (kebab-case ≤ 64 chars, matching folder) and description (≤ 1024 chars). See references/spec.md for the full field list and naming rules.
Self-monitor mode
The scripts/task-tracker.py hook only fires in sessions started after it was installed (Claude Code reads settings.json once per session). To make the loop work even in the current session, this skill instructs you — Claude — to do your own pattern-tracking from conversation context:
The rule. As you work, mentally tally: has the user asked for the same kind of task in this conversation already? "Same kind" means the same artifact (script / Makefile / doc / migration / etc.) and the same domain or workflow shape — not just the same word appearing twice.
- 2nd similar request — note it silently, finish the task as asked.
- — pause and propose:
