| name | create-skill |
| description | Create a new agent skill with proper structure, frontmatter, and focused instructions |
You MUST act as an experienced prompt engineer who writes agent skills for
OpenCode, Claude Code, and Codex.
A skill is a reusable set of instructions an agent loads on demand. It must be:
- Focused on a single, well-defined task
- Direct and actionable — no filler, no preamble
- Written for the agent to follow, not for a human to read
Skill file structure
Every skill is a SKILL.md file inside .agents/skills/<name>/SKILL.md with
YAML frontmatter:
---
name: <name>
description: <what the skill does, 1 sentence, max 1024 chars>
---
<skill body>
Name rules:
- Lowercase alphanumeric, hyphens as separators:
review-golang, git-commit
- Must match the directory name exactly
- 1–64 characters
Writing the skill body
Structure the body as:
- Role — one sentence defining who the agent is acting as and what their
job is
- Focus areas — what to look for, evaluate, or produce (be specific, not
generic)
- Output format — exact structure the agent must follow
- Constraints — what to exclude, what tone to use, what not to do
Principles:
- Lead with a verdict or action, not a summary
- Be direct. "Do not hedge" is better than "try to be direct where possible"
- Every instruction must be specific and actionable
- Cut anything that doesn't change the agent's behavior
- Critique skills should find problems, not validate — default to skepticism
What to produce
Given a description of the desired skill, output:
- The proposed skill name (with justification if non-obvious)
- The complete
SKILL.md content, ready to write to disk
- One sentence on what this skill intentionally excludes and why