Skill Creator (Meta-Skill)
Name
Skill Creator (Meta-Skill)
Description
You create and maintain Agent Skills as portable, on-demand “capability packs.” You must follow the Antigravity Skill directory standard: one folder per skill, containing a SKILL.md with YAML frontmatter plus a Markdown body.
Triggers
Use this skill when the user says anything like:
- •“Create a new skill…”
- •“Add a skill for…”
- •“Make a god-mode / baseline skill set”
- •“Update my skills folder”
- •“Make a skill that does X automatically”
Instructions
0) Non-negotiables (enforced)
- •One skill = one directory:
.agent/skills/<skill-name>/SKILL.md - •
SKILL.mdmust start with YAML frontmatter including at least:- •
description(mandatory; must be very specific) - •
name(recommended; lowercase with hyphens)
- •
- •The Markdown body must include these sections:
- •Name
- •Description
- •Triggers
- •Instructions
- •Prefer progressive disclosure:
- •Keep
SKILL.mdfocused (overview + steps). - •If content grows large, create extra files in the same folder (e.g.,
references/,examples.md,scripts/), and reference them fromSKILL.md.
- •Keep
1) Choose scope + location
Default to workspace scope unless the user explicitly asks for global:
- •Workspace:
<repo>/.agent/skills/<skill-name>/SKILL.md - •Global:
~/.gemini/antigravity/skills/<skill-name>/SKILL.md
2) Name the skill (router-friendly)
- •Use kebab-case (lowercase, hyphens):
api-integration-expert - •Make the name task-specific, not vague.
3) Write a high-signal description
The router indexes mainly on description, so it must:
- •Describe what it does
- •Say when to use it
- •Include keywords the user will naturally say
Bad: “Helps with APIs”
Good: “Implements resilient API calls (timeouts, retries, exponential backoff, jitter) for OpenAI/Gemini/Claude and centralizes auth + rate limit handling.”
4) Use the standard body template
Use this exact skeleton (fill it in, don’t freestyle it):
md
--- name: <skill-name> description: <specific behavior + when to use> --- # <Human-friendly title> ## Name <same as title> ## Description <1–3 paragraphs explaining the capability and boundaries> ## Triggers - <trigger phrase 1> - <trigger phrase 2> - ... ## Instructions ### Goal <what success looks like> ### Workflow 1) ... 2) ... 3) ... ### Constraints - ... - ... ### Examples (optional but recommended) - **Input:** ... **Output:** ...
5) Safety + repo hygiene rules
- •Never instruct the agent to exfiltrate secrets or print API keys.
- •Prefer minimal diffs; don’t do “drive-by refactors.”
- •If a skill touches destructive operations (db, deploy), include explicit constraints and “safe mode” defaults.
6) Output contract when creating skills
When you generate a new skill, output:
- •The file path
- •The full
SKILL.mdcontent - •Any additional files (if needed) with clear paths
- •A quick note on what phrases should trigger it (to validate the description)