name: skill-creator description: Create, validate, or update skills. Triggers on new skill requests, skill audits, extending kit capabilities, or packaging skills. version: 1.0.0
Skill Creator
This skill provides the methodology and standards for building new skills in the Antigravity Marketing Kit. It is the foundational skill — every other skill in this kit should be created or validated using this process.
About Skills in Antigravity
Skills are self-contained instruction packages that give Gemini specialized expertise for specific marketing tasks. Unlike multi-agent systems, skills here serve a single agent. They are context modules: when activated, Gemini reads the SKILL.md and gains procedural knowledge for that domain.
Skills are NOT documentation. They are action-oriented instructions that teach Gemini how to perform tasks, not what concepts mean.
What a skill provides:
- Trigger conditions — when to activate this skill
- Domain frameworks — proven methodologies for the task
- Step-by-step workflows — how to execute common scenarios
- Scripts — reusable code for deterministic operations
- References — detailed data too large for SKILL.md
- Assets — templates and boilerplate for output
Creating a New Skill
Step 1: Define the Scope
Before writing anything, answer these questions:
- What task does this skill enable? (Specific, not vague)
- When should Gemini activate this skill? (Trigger phrases, keywords, scenarios)
- What does the user expect as output? (File type, format, location)
- Does an existing skill already cover this? (Check
.agent/skills/— consolidate if overlap exists) - What reusable resources would help? (Scripts, reference data, templates)
Consolidation rule: Related topics belong in one skill. For example, Google Ads + Meta Ads = ads-management, not two separate skills.
Step 2: Plan the Skill Contents
Analyze concrete use cases and determine what resources to include:
Scripts (scripts/) — For repetitive, deterministic operations that benefit from code rather than natural language instructions. Scripts execute without loading into context, saving tokens.
Examples: data validation, file generation, API calls, batch processing.
References (references/) — For detailed domain knowledge that exceeds SKILL.md's 150-line limit. Loaded on demand when Gemini needs deeper context.
Examples: API schemas, formula tables, research data, checklists.
Assets (assets/) — For templates and boilerplate that Gemini copies and customizes.
Examples: email templates, document scaffolds, configuration files.
Step 3: Initialize the Skill Directory
Create the following structure:
.agent/skills/your-skill-name/
├── SKILL.md
├── scripts/ ← optional
├── references/ ← optional
└── assets/ ← optional
Naming rules for your-skill-name:
- Kebab-case only (lowercase + hyphens)
- Descriptive but concise (max 40 characters)
- Must match the
namefield in SKILL.md frontmatter exactly
Step 4: Write SKILL.md
Required format:
---
name: your-skill-name
description: Specific trigger conditions. Under 200 characters. Include keywords users would say.
---
# Skill Title
[1-2 sentence purpose statement. What this skill enables.]
## Initial Assessment
Before executing, determine:
1. [Context question relevant to this domain]
2. [Constraint or scope question]
3. [Current state / baseline question]
## Core Methodology
### [Framework or Approach Name]
[Actionable instructions, not theory]
### [Second Framework if applicable]
[Instructions]
## Workflow
### Common Scenario 1: [Name]
1. [Step with specific action]
2. [Step referencing scripts/references if applicable]
3. [Step for output]
### Common Scenario 2: [Name]
1. [Steps]
## Output Specification
- Format: [what file type / structure]
- Location: [where in assets/]
- Naming: [convention]
## Quality Checklist
- [ ] [Domain-specific quality check]
- [ ] [Another check]
- [ ] Brand context loaded from docs/brand-guidelines.md
- [ ] Output saved to assets/ with descriptive filename
Writing style rules:
- Imperative form: "Generate the headline" not "You should generate the headline"
- Verb-first instructions: "To optimize the subject line, apply..." not "The subject line can be optimized by..."
- Concise: no filler words, no restating what Gemini already knows
- Practical: every sentence should change how Gemini acts on a task
Step 5: Write Supporting Resources
For scripts:
- Python or Node.js preferred (cross-platform)
- Include error handling and clear output messages
- Respect
.envfile hierarchy:process.env>skill/.env>.agent/skills/.env - Include a requirements.txt or package.json if dependencies exist
- Test manually with real data before committing
For references:
- Each file under 150 lines; split larger docs by logical boundary
- Kebab-case filenames:
api-endpoints-auth.md, notAPIEndpoints.md - Include grep patterns in SKILL.md so Gemini knows which reference to load
For assets:
- Use
{{placeholder}}syntax for values that vary per user/project - No hardcoded brand values
Step 6: Validate the Skill
Run through this checklist before considering the skill complete:
Structure:
-
SKILL.mdexists with valid YAML frontmatter -
namefield is kebab-case and matches directory name -
descriptionis under 200 characters with specific trigger words - SKILL.md is under 150 lines
- Each reference file is under 150 lines
- No information duplicated between SKILL.md and references
- Unused example files deleted
Quality:
- Instructions use imperative form, not second person
- Teaches how to perform tasks, not what concepts mean
- Based on real marketing workflows
- Includes concrete trigger phrases and scenarios
- Cross-references related skills where applicable
- No duplication with existing skills
Scripts (if applicable):
- Cross-platform compatible
- Tested with real inputs
- Dependencies documented
- Error handling present
Step 7: Iterate
After using the skill on real tasks:
- Note where Gemini struggled or produced suboptimal output
- Identify missing context, unclear instructions, or wrong frameworks
- Update SKILL.md or add references to address gaps
- Re-test and repeat
Skill Pattern Templates
Choose the structure that best fits your skill's purpose:
Workflow-Based — For sequential processes (email campaigns, content creation)
## Overview → ## Workflow Decision Tree → ## Phase 1 → ## Phase 2 → ## Output
Task-Based — For tool collections (SEO has audit, keywords, technical checks)
## Overview → ## Quick Start → ## Task: Audit → ## Task: Keywords → ## Task: Technical
Reference/Guidelines — For standards (brand guidelines, design systems)
## Overview → ## Guidelines → ## Specifications → ## Usage Examples
Framework-Based — For strategy skills (pricing, psychology, competitor analysis)
## Overview → ## Assessment → ## Framework 1 → ## Framework 2 → ## Application
Existing Skills Catalog
Before creating a new skill, verify it doesn't overlap with these existing skills. If overlap exists, extend the existing skill instead.
| Skill | Domain |
|---|---|
brand-guidelines |
Brand context, voice, visual identity |
copywriting |
Writing frameworks, persuasion, headlines |
content-marketing |
Blog posts, editorial calendar, content audit |
email-marketing |
Sequences, templates, deliverability |
social-media |
Platform-specific content creation |
seo-optimization |
Keywords, technical SEO, programmatic SEO |
ads-management |
Google Ads, Meta Ads campaigns |
video-production |
Storyboards, scripts, production workflow |
design-system |
Design tokens, slides, visual system |
ab-test-setup |
Experiment design, statistical rigor |
marketing-psychology |
Behavioral triggers, persuasion frameworks |
pricing-strategy |
Pricing models, packaging, psychology |
competitor-analysis |
Positioning, alternatives, market gaps |
launch-strategy |
Product/campaign launch planning |
analytics |
GA4, tracking, measurement |
form-cro |
Form optimization, friction reduction |
onboarding-cro |
User onboarding, activation optimization |
ai-multimodal |
Image generation, batch processing |
References
- Validation checklist:
references/validation-checklist.md - Token efficiency guide:
references/token-efficiency.md - Script standards:
references/script-standards.md