write-skill

スター1

フォーク0

更新日2026年5月14日 14:37

Create or restructure skills for agent crews. Use when creating a new skill, splitting a skill that's too large, or converting a single-file skill to multi-file.

インストール

Codex または Claude でインストールこの Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。

ソース

smileynet

smileynet/agent-crews

GitHub リポジトリを開く Creator のリポジトリを見る

ダウンロード

Writing Skills

Process

Determine the skill type (see below)
Choose single-file or multi-file based on size/domain variance
Write SKILL.md using the appropriate template from references/
Set the description as a specific trigger (keywords + "Use when...")
Add to agent's resources in crew.yaml
Run just build

Skill Types

Type	Purpose	Template
Process	Step-by-step workflow	`references/template-process.md`
Decision	Choose between options	`references/template-decision.md`
Reference	Lookup tables, schemas	`references/template-reference.md`
Review	Evaluate/audit something	`references/template-review.md`

Single-File vs Multi-File

Single-file (shared/skills/my-skill.md): Universal, under 100 lines, no domain variants.

Multi-file (shared/skills/my-skill/):

my-skill/
├── SKILL.md              # Core (≤100 lines, loaded on trigger)
└── references/           # On-demand (agent reads when instructed)
    ├── examples-*.md     # Domain-specific examples
    └── checklist.md      # Verification/audit lists

Rules

SKILL.md = process (what to DO). References = context (what to KNOW).
Description is the trigger — specific keywords, not vague topics
Templates direct attention — even for "known" things, they anchor priority
Anti-patterns live WITH their skill, not standalone
Troubleshooting lives IN the skill (symptom→fix table) for executable skills — not as a separate file
Explicit file loading: "Read references/checklist.md" — never assume agent finds files
Conditional loading by domain: "If Rust project, read references/examples-rust.md"
Reference skills that serve multiple other skills → keep as standalone lookup (thin SKILL.md + fat references/)

Anti-Patterns

SKILL.md as knowledge dump (mix of process + encyclopedia)
Vague triggers ("help with code")
Output templates longer than instructions
Stack-specific boilerplate inline instead of in references/
Standalone "anti-patterns" skill instead of integrating with the relevant skill

このリポジトリの他の Skills

同じリポジトリ

adr-authoring

smileynet/agent-crews

Write Architecture Decision Records when making significant technical choices. Use when selecting tools, patterns, frameworks, or making structural changes that future developers need to understand.

2026-05-171

agents-md-authoring

smileynet/agent-crews

Write AGENTS.md files that serve as operating contracts for AI agents. Use when creating or updating AGENTS.md for a project. Trigger terms: AGENTS.md, agent instructions, agent contract, agent configuration, coding agent setup.

2026-05-171

changelog-discipline

smileynet/agent-crews

Quality rules for changelog entries. Use when validating, writing, or reviewing changelog content to ensure user-facing value communication.

2026-05-171

completion-protocol

smileynet/agent-crews

Task completion sequence including verification, git, signaling, and handoff. Use when finishing a task or reporting status.

2026-05-171

diagrams

smileynet/agent-crews

Architecture diagrams for documentation. Covers ASCII, Mermaid, C4, and D2. Applies C4 methodology by default. Trigger terms: diagram, flowchart, architecture diagram, sequence diagram, C4, mermaid, ascii diagram.

2026-05-171

diataxis-classification

smileynet/agent-crews

Classify documentation by audience and content type. Use when planning, auditing, or placing project documentation.

2026-05-171

Writing Skills

Process

Determine the skill type (see below)

Choose single-file or multi-file based on size/domain variance

Write SKILL.md using the appropriate template from references/

Set the description as a specific trigger (keywords + "Use when...")

Add to agent's resources in crew.yaml

Run just build

Skill Types

Type

Purpose

Template

Process

Step-by-step workflow

references/template-process.md

Decision

Choose between options

references/template-decision.md

Reference

Lookup tables, schemas

references/template-reference.md

Review

Evaluate/audit something

references/template-review.md

Single-File vs Multi-File

Single-file (shared/skills/my-skill.md): Universal, under 100 lines, no domain variants.

Multi-file (shared/skills/my-skill/):

my-skill/ ├── SKILL.md # Core (≤100 lines, loaded on trigger) └── references/ # On-demand (agent reads when instructed) ├── examples-*.md # Domain-specific examples └── checklist.md # Verification/audit lists

Rules

SKILL.md = process (what to DO). References = context (what to KNOW).

Description is the trigger — specific keywords, not vague topics

Templates direct attention — even for "known" things, they anchor priority

Anti-patterns live WITH their skill, not standalone

Troubleshooting lives IN the skill (symptom→fix table) for executable skills — not as a separate file

Explicit file loading: "Read references/checklist.md" — never assume agent finds files

Conditional loading by domain: "If Rust project, read references/examples-rust.md"

Reference skills that serve multiple other skills → keep as standalone lookup (thin SKILL.md + fat references/)

Anti-Patterns

SKILL.md as knowledge dump (mix of process + encyclopedia)

Vague triggers ("help with code")

Output templates longer than instructions

Stack-specific boilerplate inline instead of in references/

Standalone "anti-patterns" skill instead of integrating with the relevant skill

name	write-skill
description	Create or restructure skills for agent crews. Use when creating a new skill, splitting a skill that's too large, or converting a single-file skill to multi-file.