BIGboss248/nextjs-wordpress

Antigravity Customizer (ag-customizer)

Use this skill to autonomously design, scaffold, configure, migrate, and validate customizations for Google Antigravity across workspace and global scopes.

---

Quick Navigation Decision Matrix

Determine which customization mechanism best solves the problem:

Need	Mechanism	File Location	Reference Guide
Teach the agent a multi-step runbook, procedure, or slash command	Skill	.agents/skills/<name>/SKILL.md or ~/.gemini/antigravity/skills/<name>/SKILL.md	skills-guide.md
Enforce persistent constraints, coding standards, or style rules	Rule	.agents/rules/<name>.md, AGENTS.md, or ~/.gemini/GEMINI.md	rules-guide.md
Connect external services, local databases, or custom tools	MCP Server	.agents/mcp_config.json or ~/.gemini/config/mcp_config.json	mcp-guide.md
Intercept lifecycle events (gate commands, inject prompts, stop watchdog)	Lifecycle Hook	.agents/hooks.json or ~/.gemini/config/hooks.json	hooks-guide.md
Bundle skills, rules, hooks, and MCP into a shareable package	Plugin	.agents/plugins/<name>/ or ~/.gemini/config/plugins/<name>/	plugins-and-configs.md
Register non-standard paths or share customizations via VCS	JSON Config	.agents/skills.json or .agents/plugins.json	plugins-and-configs.md
Upgrade deprecated workflows into modern slash-command skills	Workflow Migration	.agents/workflows/*.md -> .agents/skills/<name>/SKILL.md	workflows-migration.md

---

Operational Workflows

1. Adding a New Skill

1. Determine Scope:
   • Workspace: .agents/skills/<name>/ (checked into repository for team use).
   • Global: ~/.gemini/antigravity/skills/<name>/ (available across all projects).
2. Structure the Directory:

   <customization_root>/skills/<skill-name>/
   ├── SKILL.md             # Required: Entry point with YAML frontmatter
   ├── scripts/             # Optional: Helper scripts (run with --help pattern)
   ├── references/          # Optional: Disclosed manuals & heavy documentation
   └── examples/            # Optional: Sample code or output references
   

3. Draft Frontmatter:
   • name: Lowercase, hyphenated (e.g. nextjs-otel, docker-deploy). This also registers as a slash command (/<name>).
   • description: 1-3 sentences in third person, explicitly stating what the skill does and specific trigger phrases/scenarios when the agent should activate it.
4. Apply Progressive Disclosure:
   • Keep the main SKILL.md tightly focused on the sequence of operational steps and completion criteria.
   • Disclose deep reference material into references/<topic>.md files and link them relatively.
   • Refer to skills-guide.md for structure templates.
5. Obsidian Mirroring Check:
   • When updating or creating skills in this workspace, mirror the skill folder to d:/Scripts/Obsidian/skills/<name>/ per repository guidelines.

---

2. Creating a Rule

1. Determine Scope & Format:
   • Workspace Rule: .agents/rules/<rule-name>.md or AGENTS.md at project root.
   • Global Rule: ~/.gemini/GEMINI.md.
2. Choose Activation Mode (for .agents/rules/*.md):
   • always_on: Injected unconditionally into every turn. Use sparingly to preserve context.
   • model_decision: Disclosed when the agent determines relevance based on the description.
   • glob: Automatically activated whenever matching files are touched or opened (e.g. *.tsx, src/api/**/*.ts).
   • manual: Only loaded when explicitly invoked via @mention in chat (e.g. @my-rule.md).
3. Author Content:
   • Keep within the 12,000-character budget per file.
   • State target behaviors in the positive ("Do X", "Always structure Y") rather than negative prohibitions.
   • Use @filename to link to companion context files if needed.
   • Refer to rules-guide.md for full syntax and activation frontmatter.

---

3. Configuring an MCP Server

1. Determine File Location:
   • Workspace: <workspace>/.agents/mcp_config.json
   • Global: ~/.gemini/config/mcp_config.json
2. Select Transport:
   • stdio (Local executable):

     {
       "mcpServers": {
         "my-server": {
           "command": "npx",
           "args": ["-y", "@modelcontextprotocol/server-sqlite", "app.db"],
           "env": { "SQLITE_READONLY": "true" }
         }
       }
     }
     

   • Remote (SSE / Streamable HTTP):

     [!IMPORTANT] Always use the key serverUrl. The legacy keys url or httpUrl are not supported.

     {
       "mcpServers": {
         "my-remote": {
           "serverUrl": "https://api.example.com/mcp/",
           "headers": { "Authorization": "Bearer YOUR_TOKEN" }
         }
       }
     }
     

3. Authentication:
   • Google Application Default Credentials: "authProviderType": "google_credentials" (requires gcloud auth application-default login).
   • OAuth: "oauth": { "clientId": "...", "clientSecret": "..." } with redirect URI https://antigravity.google/oauth-callback.
4. Tool Filtering:
   • Temporarily disable tools without removing server: "disabledTools": ["destructive_command"].
   • Refer to mcp-guide.md for pre-built configurations.

---

4. Setting Up Lifecycle Hooks

1. Target File: .agents/hooks.json (or ~/.gemini/config/hooks.json).
2. Supported Events:
   • PreToolUse: Match tool names (run_command, write_to_file, *) to gate execution (allow, deny, ask, force_ask) or shallow-overwrite arguments.
   • PostToolUse: Audit output, run linters, or format code after tool completes.
   • PreInvocation: Inject context or system instructions (injectSteps with ephemeralMessage) before model processes the turn.
   • PostInvocation: Force continuation (terminationBehavior: "force_continue") or stop.
   • Stop: Inspect task completion (terminationReason, fullyIdle) and return decision: "continue" with reason to block premature exit.
3. Contract Rules:
   • Execution passes context via stdin and receives response via stdout.
   • All JSON payload keys use camelCase (conversationId, stepIdx, toolCall, etc.).
   • Commands execute synchronously and block the agent loop until timeout (default 30s).
   • Refer to hooks-guide.md for templates in Python, PowerShell, and Bash.

---

5. Packaging a Plugin or VCS Shared Config

1. Plugin Package:
   • Group related skills, rules, hooks, and MCP servers into .agents/plugins/<plugin-name>/.
   • Add plugin.json manifest ({"name": "my-plugin"}).
   • Manage activation state in config.json under "plugins": { "<plugin-name>": { "enabled": true } }.
2. JSON Configurations (skills.json / plugins.json):
   • Register external directories or shared team repositories.
   • Support entries (paths) and inherits (other configs) with include_only and exclude regex filters.
   • Refer to plugins-and-configs.md for complete manifest examples.

---

6. Modernizing Legacy Workflows

1. Inspect for deprecated .agents/workflows/*.md or ~/.gemini/config/workflows/*.md.
2. Extract operational instructions and generate equivalent SKILL.md with proper YAML frontmatter.
3. Archive original file to <name>.md.bak (safe preservation).
4. Refer to workflows-migration.md for the automated migration runbook.

---

Validation & Verification Checklist

Before considering any customization complete:

• JSON Syntax: Ensure all .json files parse cleanly without syntax errors or comments.
• Frontmatter Check: Verify SKILL.md contains valid YAML frontmatter with name and descriptive description.
• Character Budget: Ensure rule Markdown files are within the 12,000 character limit.
• MCP Connection: Verify command binaries or remote URLs exist and authenticate.
• Hook Contract: Verify hook scripts accept camelCase JSON on stdin and return valid JSON on stdout.
• Obsidian Mirroring: For skills in this workspace, sync the created skill folder to d:/Scripts/Obsidian/skills/<name>/.