| name | writing-skills |
| description | Meta-skill for creating new skills in oh-my-colab format. Use when RETRO identifies a pattern worth formalizing. Ensures correct folder structure, 200-line discipline, proper frontmatter, and token-efficient design.
|
| allowed-tools | Read Write |
Writing Skills
Use after RETRO identifies a pattern worth formalizing as a reusable skill.
Skill Folder Structure
skills/{skill-name}/
├── SKILL.md ← core instructions, ≤200 lines
├── references/ ← detailed docs, loaded on demand
│ └── *.md
├── scripts/ ← executable tools, only output enters context
│ └── *.sh / *.py
└── assets/ ← templates, blank forms
└── *.md
Required SKILL.md Frontmatter
---
name: skill-name
description: >
When this skill activates. What it produces. Key trigger words or contexts.
Make this searchable. Be specific about when to use vs not use.
allowed-tools: Read Write Bash
---
The 200-Line Rule
SKILL.md must stay under 200 lines. If content exceeds it:
- Identify the long section (usually examples or detailed reference)
- Move it to
references/{descriptive-name}.md
- Replace with:
See references/{name}.md for [brief description]
What stays in SKILL.md: the protocol (what to do, in what order)
What goes to references/: examples, full checklists, edge cases, templates
What goes to scripts/: validation, detection, generation logic
What goes to assets/: blank templates the user copies
Writing Good Descriptions
The description is ALL the agent reads before deciding to load this skill.
Bad: "A skill for debugging code."
Good: "Hypothesis-driven debugging. Triggers on debug, error, broken, failing.
Forces explicit hypothesis before fix. Prevents guess-and-check debugging."
Promoting from Draft
Draft skills go to .ohc/skills/{name}.md (single file, informal).
Promoted skills go to skills/{name}/SKILL.md (full folder structure).
ohc skill draft {name}
ohc skill promote {name}
See references/skill-spec.md for the complete Anthropic skill specification.
See references/frontmatter-guide.md for all frontmatter fields.