// This skill should be used when the user asks to "create a skill", "forge a skill", "generate a skill", "build a Claude skill", "make a new skill", or wants to understand skill creation methodology. Provides research-grounded methodology for creating high-quality Claude Code skills with official documentation verification.
| name | forge-methodology |
| description | This skill should be used when the user asks to "create a skill", "forge a skill", "generate a skill", "build a Claude skill", "make a new skill", or wants to understand skill creation methodology. Provides research-grounded methodology for creating high-quality Claude Code skills with official documentation verification. |
| version | 1.0.0 |
Create research-grounded, high-quality Claude Code skills by following a rigorous methodology that verifies all information against official documentation before generation.
NEVER generate skill content from training data alone. Every piece of guidance in a skill must be:
Before creating any skill, research the official skill format:
DeepWiki Query: Use mcp__deepwiki__ask_question on anthropics/claude-code for:
WebFetch Official Sources:
https://github.com/anthropics/skills — Official skills repositoryhttps://www.anthropic.com/news/skills — Skills announcementhttps://github.com/anthropics/claude-code/blob/main/plugins/README.md — Plugin documentationContext7 Library Docs: If the skill involves a library, use mcp__context7__resolve-library-id then mcp__context7__get-library-docs for accurate API information.
For the specific domain the skill covers:
Identify Authoritative Sources:
WebSearch for Current Information:
"[domain] official documentation 2024 2025""[domain] best practices site:[official-domain]"WebFetch and Extract:
Combine research into skill content:
Based on official Anthropic documentation:
---
name: skill-name-kebab-case
description: This skill should be used when the user asks to "[trigger phrase 1]", "[trigger phrase 2]", "[trigger phrase 3]". [Brief description of capability].
version: 1.0.0
---
The description field is critical — it determines when Claude activates the skill.
Must include:
Good example:
This skill should be used when the user asks to "deploy to Kubernetes", "create K8s manifests", "configure kubectl", or mentions Kubernetes deployment patterns.
Bad example:
Helps with Kubernetes tasks.
Write the SKILL.md body in imperative/infinitive form (verb-first):
Correct: "Create the deployment manifest with..." Incorrect: "You should create the deployment manifest..."
Keep SKILL.md body lean (1,500-2,000 words). Move detailed content to references/.
skill-name/
├── SKILL.md # Required: Core instructions (lean)
├── references/ # Detailed documentation
│ ├── patterns.md # Common patterns and examples
│ ├── api-reference.md # API details if applicable
│ └── sources.md # Citation URLs
├── examples/ # Working code examples
│ └── example-usage.sh
└── scripts/ # Utility scripts
└── validate.sh
Skills load in three levels to optimize token usage:
| Level | Content | When Loaded | Size Target |
|---|---|---|---|
| 1 | name + description | Always | ~100 words |
| 2 | SKILL.md body | When triggered | <2,000 words |
| 3 | references/, examples/ | As needed | Unlimited |
Keep SKILL.md lean. If content exceeds 2,000 words, move to references/.
Before finalizing any skill, verify:
Structure:
name and description fields presentDescription Quality:
Content Quality:
Accuracy:
Prioritize sources in this order:
Official documentation (highest trust)
Official GitHub repositories
Authoritative technical sources
Verified community sources (use with caution)
General web (lowest trust)
For detailed patterns and implementation guides, consult:
references/skill-patterns.md — Common skill patterns from official examplesreferences/research-guide.md — Detailed research methodologyreferences/quality-checklist.md — Comprehensive validation criteriareferences/sources.md — Official documentation URLsFrom official Anthropic guidance:
"While this makes them powerful, it also means that malicious skills may introduce vulnerabilities... We recommend installing skills only from trusted sources."
When creating skills: