Skill Creator

What I do

Create and validate OpenCode skills following the Anthropic Agent Skills spec. Skills that violate these rules fail silently: not discovered, not matched, or waste tokens.

Directory structure

.claude/skills/{skill-name}/
└── SKILL.md

One directory per skill. Directory name must match name in frontmatter. Only one file: SKILL.md (uppercase). No subdirectories.

SKILL.md format

---
name: {skill-name}
description: >-
  {capability summary}. {use cases}. {proactive triggers}.
  Examples:
  - user: "{english phrase}" -> {action}
  - user: "{spanish phrase}" -> {action}
  Never use this for {anti-pattern}.
---

## Environment
All commands run via `ssh web`.

## Instructions
{Step-by-step actionable content}

## Verification
{How to verify correctness}

Frontmatter rules

name (REQUIRED)

• 1-64 chars, lowercase alphanumeric + single hyphens
• Must match directory name exactly
• Valid: module-scaffold, quality-checks | Invalid: drupal_module, MySkill

description (REQUIRED, 1-1024 chars)

This is the ONLY text the LLM sees during skill selection. Structure:

1. Capability: What it does (1 sentence)
2. Use cases: When to invoke (1-2 sentences)
3. Examples: 2-4 with user: "X" -> Y format
4. Anti-pattern: "Never use for X"
5. Must include Spanish AND English triggers

Progressive disclosure

Startup: Only name + description loaded (XML summary)
User message -> LLM matches description -> skill tool called -> full body loaded

If description doesn't match user intent, body NEVER loads.

Validation checklist

• [ ] Directory name = name field
• [ ] Name: lowercase, hyphens only, 1-64 chars
• [ ] Description: 1-1024 chars
• [ ] Description has: capability, use cases, 2+ examples, anti-pattern
• [ ] Description has Spanish triggers
• [ ] YAML frontmatter valid
• [ ] Body has actionable instructions (not prose)
• [ ] Commands use ssh web
• [ ] No duplication with other skills or rules/
• [ ] No conflicts with CLAUDE.md policies
• [ ] File named exactly SKILL.md
• [ ] Body under 200 lines

Audit mode

When reviewing existing skills:

find .claude/skills -name "SKILL.md" | while read f; do
  dir=$(basename $(dirname "$f"))
  name=$(grep "^name:" "$f" | head -1 | awk '{print $2}')
  lines=$(wc -l ` |
| No anti-pattern | Add "Never use for X" |
| Bare commands | Prefix `ssh web` |
| Body 200+ lines | Split into skill + rule reference |
| English-only triggers | Add Spanish phrases |
| Duplicates other skill | Merge or add anti-pattern boundary |

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [trebormc](https://github.com/trebormc)
- **Source:** [trebormc/drupal-ai-agents](https://github.com/trebormc/drupal-ai-agents)
- **License:** Apache-2.0

Install and usage instructions live in the source repository linked above.