| name | skill-creator |
| description | Guide for creating effective skills in .github/skills/. Use when creating a new skill (or updating an existing skill) that extends Agent's capabilities with specialized knowledge, workflows, or tool integrations. |
Skill Creator
This skill provides guidance for creating effective skills for this repository.
About Skills
Skills are modular, self-contained packages that extend Agent's capabilities by providing specialized knowledge, workflows, and tools. Think of them as "onboarding guides" for specific domains or tasks.
Project skills live in .github/skills/<skill-name>/SKILL.md.
Core Principles
Concise is Key
The context window is a public good. Only add context Agent doesn't already have. Challenge each piece: "Does Agent really need this explanation?" and "Does this paragraph justify its token cost?"
Prefer concise examples over verbose explanations.
Set Appropriate Degrees of Freedom
- High freedom (text instructions): When multiple approaches are valid or decisions depend on context.
- Medium freedom (pseudocode with parameters): When a preferred pattern exists but some variation is acceptable.
- Low freedom (specific scripts, few parameters): When operations are fragile, error-prone, or must follow an exact sequence.
Anatomy of a Skill
.github/skills/<skill-name>/
├── SKILL.md (required)
│ ├── YAML frontmatter: name, description (required)
│ └── Markdown instructions (required)
└── references/ (optional — loaded only when needed)
SKILL.md Frontmatter
Required fields:
---
name: <skill-name>
description: <one-line description used for auto-discovery and invocation>
---
The description is the primary discovery surface — it must clearly state when the skill triggers. Include exact trigger phrases if the skill should auto-invoke.
References (optional)
Documentation intended to be loaded into context as needed. Use for:
- Detailed workflow guides
- API or protocol specifications relevant to the project
- Domain-specific reference material too large for SKILL.md
Keep SKILL.md under ~500 lines. Move detail into references/ files and link them from SKILL.md.
What NOT to include
Do not create:
README.md
INSTALLATION_GUIDE.md
CHANGELOG.md
- Any auxiliary docs about the skill creation process itself
The skill should contain only information needed for an agent to do the job at hand.
Progressive Disclosure
Skills use a three-level loading system:
- Metadata (name + description) — Always in context.
- SKILL.md body — Loaded when skill triggers.
- References — Loaded only as needed by Agent.
Keep SKILL.md lean. Move detail to references/ files and reference them explicitly.
Writing a New Skill
- Pick a short, lowercase, hyphenated name matching the domain (e.g.,
backend-bug-review).
- Write a
description that includes exact trigger phrases users or the agent will use.
- Write directive, action-oriented instructions — record rules, not narratives.
- Tie guidance to actual files and conventions in this repository (
api/, worker/, node:test).
- Avoid duplicating rules that already exist in another skill — link instead.
- Place the file at
.github/skills/<skill-name>/SKILL.md.
Updating an Existing Skill
Prefer updating an existing skill over creating a new one when a clear owner exists. See skill-updater for the update workflow.