Skill Composition
Compose, split, test, and sandbox preview skills
composetestsplitsandbox
compose#
Combine multiple skills into a single "super-skill."
skills compose <skills...> [options]
| Option | Description |
|---|---|
-o, --output <name> | Output skill name (default: composed-skill) |
-s, --strategy <type> | Composition strategy: merge, chain, conditional (default: merge) |
--no-dedup | Disable deduplication in merge mode |
--save | Auto-save to global skills directory |
Composition strategies:
| Strategy | Description |
|---|---|
merge | Combines all sections, deduplicates similar lines (default) |
chain | Sequences skills with ordered phase markers |
conditional | Wraps each skill in a conditional activation block |
Examples:
# Merge two frontend skills
skills compose frontend-design frontend-code-review -o frontend-pro
# Chain skills in sequence
skills compose setup coding-style testing -o full-workflow -s chain --save
# Conditional activation
skills compose python-pro typescript-pro -o polyglot -s conditional
Output:
✔ Composed "combined-frontend"
Source skills: frontend-design, frontend-code-review
Strategy: merge
Token count: 1729
Deduplicated: 1 redundant lines removed
Saved to: ~/.antigravity/skills/combined-frontend/SKILL.md
test#
Run quality tests and assertions against skills.
skills test [skills...] [options]
| Option | Description |
|---|---|
-a, --all | Test all installed skills |
--json | Output results as JSON |
-v, --verbose | Show all assertions (not just failures) |
10 built-in assertions:
| Category | Assertion | Description |
|---|---|---|
| Structure | SKILL.md exists | Skill file must be present |
| Structure | Valid frontmatter | Must parse without errors |
| Structure | Has name | Frontmatter includes name field |
| Structure | Has description | Frontmatter includes description field |
| Content | Has sections | At least 2 markdown headings |
| Content | Has code examples | At least one fenced code block |
| Content | Has "when to use" | Describes activation criteria |
| Quality | Description < 200 chars | Concise, scannable description |
| Quality | No TODOs/FIXMEs | No placeholder content |
| Quality | Minimum length | Content is at least 100 characters |
Custom tests: Create a skill-test.yml in your skill directory:
tests:
- name: "Uses TypeScript examples"
type: contains
value: "typescript"
- name: "Mentions error handling"
type: contains
value: "error"
Examples:
skills test frontend-design # Test a specific skill
skills test --all --verbose # Test all, show all assertions
skills test --all --json # Machine-readable for CI
Output:
🧪 Skill Test Results
✗ frontend-design 70% (7ms)
✗ Has "when to use" section
✗ Has code examples
✗ Description is concise
Summary: 0 passed / 1 failed / 1 total
CI Integration: Exits with code 1 on any failure, making it ideal for CI pipelines.
split#
Split a large skill into focused sub-skills by topic.
skills split <skill> [options]
| Option | Description |
|---|---|
-m, --min-sections <n> | Minimum sections per sub-skill (default: 2) |
--save <dir> | Save sub-skills to directory |
--dry-run | Only preview the split, do not save |
--json | Output as JSON |
Topic categories used for clustering:
setup, coding-style, testing, architecture, deployment, security, api, database, documentation, performance
Examples:
skills split skill-creator --dry-run # Preview the split
skills split big-skill --save ./output # Save sub-skills
skills split monolith --min-sections 3 # Require 3+ sections per group
Output:
✂️ Skill Splitter: skill-creator
Original: 4426 tokens
Can be split into 8 sub-skills:
📦 testing (300 tokens) — About Skills, What Skills Provide, Iterate
📦 deployment (785 tokens) — Core Principles, Concise is Key, ...
📦 setup (955 tokens) — SKILL.md (required), Quick start, ...
📦 database (983 tokens) — Bundled Resources, Planning, ...
...
sandbox#
Preview a skill's quality, conflicts, and token impact before installing.
skills sandbox <source> [options]
| Option | Description |
|---|---|
-f, --format <format> | Output format: text, json |
Accepts local paths or remote GitHub sources (@owner/repo). Runs three analyses:
- •Skill Info — Name, description, token count, line count
- •Quality Score — Grade (A–F) from the built-in test framework
- •Conflict Analysis — Checks for overlaps with already-installed skills
Examples:
skills sandbox ~/.antigravity/skills/my-skill # Local skill
skills sandbox @owner/repo # Remote skill
skills sandbox ./my-skill # Relative path
Output:
🧪 Sandbox Preview: frontend-design
📋 Skill Info
Name: frontend-design
Tokens: 1069
Lines: 42
🧪 Quality Score
Grade: C (70%)
Issues:
✗ Has "when to use" section
✗ Has code examples
⚔️ Conflict Analysis
1 overlap(s) found (~842 wasted tokens)
📊 Verdict
This skill would consume ~1069 tokens of your context budget.