| name | ecc-skill-development-guide |
| description | Use when writing a new skill, reviewing a skill, refactoring an existing one, or onboarding someone to the skill ecosystem. Triggers: "how do I write a good skill", "review this skill", "skill format", "skill frontmatter", "skill description contract", "skill length budget", "skill anti-patterns", "skill review checklist". |
Skill Development Guide
A comprehensive guide to creating effective skills. This is the meta-skill — read it before writing your first skill, and re-read it before reviewing someone else's.
What are skills?
Skills are knowledge modules that an agent (Copilot, Claude Code, Codex, OpenClaw, etc.) loads based on context. They are pure instructions, not executable code — the agent reads the SKILL.md, follows it, and uses the same tools it already has. Skills can include helper scripts that the agent invokes, but the skill itself is documentation, not software.
File structure
skills/
└── <skill-name>/
├── SKILL.md # Required. <500 lines.
├── examples/ # Optional. Concrete worked examples.
│ ├── basic.md
│ └── advanced.md
└── references/ # Optional. External links, jargon, deep dives.
└── links.md
SKILL.md format
---
name: skill-name
description: Brief description shown in the skill list and used for auto-activation.
origin: ECC # or community, or the source repo
version: "1.0.0"
---
# Skill Title
Brief overview of what this skill covers.
## When to Activate
Describe scenarios where the agent should use this skill. This is the load-bearing section
for auto-activation — if it's vague, the skill gets loaded at the wrong times.
## Core Concepts
Main patterns and guidelines. Use the project's domain glossary vocabulary, not invented
synonyms.
## Code Examples
```typescript
// Include practical, tested examples
function example() {
// Well-commented code
}
Anti-Patterns
Show what NOT to do, with concrete examples. A good anti-pattern section prevents the same
mistake across many agents.
Best Practices
- Actionable guidelines
- Do's and don'ts
- Common pitfalls to avoid
Related Skills
Link to complementary skills (e.g. related-skill-1, related-skill-2).
## Frontmatter fields
| Field | Required | Description |
|--------------|----------|-----------------------------------------------------------------------------|
| `name` | Yes | Lowercase, hyphenated identifier (e.g. `react-patterns`). Must match the directory name. |
| `description`| Yes | One-line description for the skill list and auto-activation. |
| `origin` | No | Source identifier (e.g. `ECC`, `community`, your repo name). |
| `tags` | No | Array of tags for categorisation. |
| `version` | No | Skill version for tracking updates. |
| `license` | No | SPDX identifier (e.g. `MIT`). Default to repo's license. |
| `metadata` | No | Free-form key-value for author/source/links. |
## Description is the most important field
The description is the *only* thing the agent sees when deciding whether to load the skill. It is surfaced in the system prompt alongside every other installed skill.
**Goal:** give the agent just enough info to know:
1. What capability this skill provides.
2. When / why to trigger it (specific keywords, contexts, file types).
**Format:**
- Max 1024 chars.
- Third person.
- First sentence: what it does.
- Second sentence: "Use when [specific triggers]."
**Good example:**
> Discipline the diagnosis loop for hard bugs: build a feedback loop, hypothesise, instrument, fix, regression-test. Use when the user reports a bug, says "debug this", describes a performance regression, or asks for "diagnose".
**Bad example:**
> Helps fix bugs.
The bad example gives the agent no signal to distinguish it from generic "fix" skills.
## Skill categories
| Category | Purpose | Examples |
|------------------------|----------------------------------------------------------|------------------------------------------|
| **Language Standards** | Idioms, conventions, best practices | `python-patterns`, `golang-patterns` |
| **Framework Patterns** | Framework-specific guidance | `django-patterns`, `nextjs-patterns` |
| **Workflow** | Step-by-step processes | `tdd-workflow`, `refactoring-workflow` |
| **Domain Knowledge** | Specialised domains | `security-review`, `api-design` |
| **Tool Integration** | Tool / library / service usage | `docker-patterns`, `supabase-patterns` |
| **Template** | Project-specific skill template | `docs/examples/project-guidelines-template.md` |
## Writing effective skill content
- **Be specific.** "Be careful with X" is a wish. "X fails when Y because Z. Fix by doing W" is a skill.
- **Show, don't tell.** Code examples beat prose. Worked examples beat abstract examples.
- **Anchor in the project's language.** Use the terms the project already uses. If the project's glossary says "Order intake," don't say "the place where orders are processed."
- **Include the anti-pattern.** A good "Don't do X because it broke in 2024-Q1" is worth 100 words of best-practice prose.
- **Length budget.** 500 lines for `SKILL.md` is a hard ceiling. If you need more, split into `REFERENCE.md` (deep dive) or `EXAMPLES.md` (worked examples).
- **No time-sensitive info.** "As of 2024" is a smell. Use a versioned script for things that go stale.
## Best practices
- Test the skill against the actual agent. The description is the contract; if the agent loads the wrong skill, the description is wrong.
- Keep examples realistic. Abstracted examples teach abstract lessons. Real examples with real names teach real lessons.
- Update the skill's `version` when you materially change it. Track changes in a `CHANGELOG.md` if the skill has downstream users.
- Be ruthless about cutting. If a section doesn't earn its keep when the agent reads it, delete it.
## Common anti-patterns
- **Vague descriptions** — "Helps with X" doesn't tell the agent when to load the skill.
- **Overloaded skills** — one skill doing five things. Split by what the user would search for.
- **No anti-patterns section** — best-practices-only skills produce "I followed the skill and still failed" situations.
- **Wall of text** — skills that read like a blog post. Skills are reference material; bullet points and code blocks beat paragraphs.
- **Skill that does the work itself** — skills are instructions, not automation. If you want a script, write a script; the skill teaches when to call it.
## Testing your skill
Before publishing:
1. **Load test** — can the agent pick this skill from its description? Run a few trigger phrases and see if it loads.
2. **Apply test** — once loaded, does the agent produce output that matches the skill's examples?
3. **Skip test** — when the trigger doesn't match, does the agent *not* load the skill? (False positives are worse than false negatives.)
4. **Edit test** — can someone else (or future-you) update the skill without breaking the description contract?
## Submitting / publishing
- Place the skill in the appropriate `skills/<name>/` directory.
- Update the parent `README.md` (if your project has one) to reference the new skill.
- Add an entry to any `manifest.json` or `plugin.json` if the agent surface requires it.
- Tag the commit. Don't bury skill changes in feature work.
## Source / license
Adapted from [mattpocock/ECC · SKILL-DEVELOPMENT-GUIDE.md](https://github.com/JZKK720/ECC), MIT.
