Skill Generator (Meta-Skill)

Goal

Create properly formatted, high-quality Antigravity skills that follow all conventions and will be reliably triggered by the agent when relevant.

Instructions

Step 1: Define the Skill

Before creating any files, define:

• Name: lowercase, hyphenated (e.g., run-tests, deploy-staging)
• Trigger description: The description that will go in the YAML frontmatter. This MUST be specific enough for the LLM to match semantically.
  • BAD: "Database tools" (too vague — agent will not know when to use this)
  • GOOD: "Executes the project test suite using pytest and reports results with coverage metrics"
• What it does: Plain-English explanation for the orchestrator
• What files it needs: Scripts? Resources? Templates?

Step 2: Create the Directory

Create the skill directory at .agent/skills/<skill-name>/:

<skill-name>/
├── SKILL.md           # Required: definition and instructions
├── scripts/           # Optional: executable scripts
└── resources/         # Optional: templates, docs, references

Step 3: Write the SKILL.md

Use the template in resources/skill-template.md. The file has two sections:

YAML Frontmatter (between --- markers):

---
name: <skill-name>
description: "<Specific trigger phrase. Describe WHEN this skill should activate and WHAT user intent it matches.>"
---

Markdown Body:

# <Skill Name>

## Goal
[One sentence: what does this skill accomplish?]

## Instructions
[Step-by-step instructions for the agent to follow when this skill activates]

## Examples
[At least one example of input → output to show the expected behavior]

## Constraints
[What this skill should NOT do. Edge cases. Limitations.]

Step 4: Create Supporting Files

• Scripts (scripts/): If the skill automates commands, create scripts that the agent can execute. Use Python over Bash for readability (the orchestrator is a non-coder).
• Resources (resources/): If the skill uses templates, reference docs, or static content, place them here. The SKILL.md should reference them by relative path.

Step 5: Quality Check

Before finalizing, verify using resources/skill-checklist.md:

• Is the description specific enough for semantic matching?
• Are instructions clear and unambiguous?
• Are edge cases covered in constraints?
• Does it follow Antigravity SKILL.md conventions?
• Will the orchestrator understand what this skill does?

Step 6: Log the Creation

Append an entry to .agent/knowledge/evolution-log.md:

[DATE] TYPE: skill NAME: <skill-name>
  TRIGGER: [What caused this skill to be created — pattern crystallization, user request, bootstrap]
  DESCRIPTION: [Plain-English description]
  METRIC: [What it improves]
  STATUS: active

Step 7: Inform the Orchestrator

Tell the orchestrator:

• What the new skill does (plain English)
• When it will trigger automatically
• How they can also trigger it manually if needed
• That they can find it at .agent/skills/<skill-name>/SKILL.md

Constraints

• Never create a skill without orchestrator approval
• Never create a skill that duplicates an existing one — check .agent/skills/ first
• Never create a skill that modifies the core Ground Zero rules (files numbered 00-06)
• Skills must always be explainable in one sentence