g-skl-platform-antigravity

name: g-skl-platform-antigravity description: Authoritative reference for Google Antigravity (agent-first IDE + CLI + SDK) customization in gald3r projects. Covers AGENTS.md/GEMINI.md instruction files, .agents/ rules/skills/hooks + .antigravity/mcp.json, Workflows (slash commands), dynamic subagents, lifecycle hooks, Scheduled Tasks, and gald3r install verification. Post-2.0-relaunch (I/O 2026); skill path + hook payload pin via install test. crawl_max_age_days: 7 vault_doc_path: research/platforms/antigravity/ vault_docs_url: https://antigravity.google/docs/home docs_url: https://antigravity.google/docs/home docs_url_secondary: - https://antigravity.google/docs/hooks - https://antigravity.google/docs/subagents - https://antigravity.google/docs/skills - https://antigravity.google/docs/rules-workflows - https://antigravity.google/docs/mcp - https://antigravity.google/docs/command last_doc_scan: 2026-06-02 capability_status: hooks: "✅ native lifecycle hooks in hooks.json (before/after_tool_call, before/after_model_call, on_loop_stop, on_error; stdin/stdout JSON; shell scripts)" rules: "✅ Markdown rules (Manual/Always On/Model Decision/Glob) in .agents/rules + GEMINI.md; reads AGENTS.md" skills: "✅ Agent Skills (Anthropic SKILL.md) in .agents/.antigravity/.agent skills dirs (path pin via install test)" commands: "✅ Workflows (saved-prompt slash commands) /workflow-name; ~/.gemini/antigravity/global_workflows/" agents: "✅ native dynamic subagents (Orchestrator-spawned); ⚠️ no file-based g-agnt-*.md discovery" mcp: "✅ native — .antigravity/mcp.json or ~/.gemini/antigravity/mcp_config.json + MCP Store GUI" token_budget: low subsystem_memberships: [PLATFORM_INTEGRATION]

g-skl-platform-antigravity

Activate for: setting up gald3r with Google Antigravity (IDE / CLI / SDK), authoring workflows/rules/skills/hooks, wiring MCP, or verifying the Antigravity gald3r install.

Full breakdown + evidence URLs in PLATFORM_SPEC.md (this folder). Status: ✅ near-full parity — the post-2.0 IDE/CLI harness natively supports Workflows (commands), rules, dynamic subagents, skills, hooks, and MCP, and reads AGENTS.md + GEMINI.md while discovering Anthropic-format SKILL.md, so gald3r's AGENTS.md + g-skl-* artifacts are largely reusable. (Verified 2026-06-02 against https://antigravity.google/docs — JS-SPA docs; evidence cross-checked from verbatim extractions + Google codelabs.)

⚠️ VOLATILE / NEW PLATFORM. Antigravity relaunched agent-first ~2026-05-19; Subagents / Hooks / Scheduled Tasks announced at I/O 2026. Keep crawl_max_age_days: 7. Two unknowns to pin via a live install test before hard-coding into installers: (1) the exact skills dir (.agents/skills vs .antigravity/skills vs .agent/skills), and (2) the hook payload schema.

1. Platform Overview

Google Antigravity (DeepMind) — an agent-first platform: a desktop IDE, a CLI, and an SDK sharing one agent harness (common settings/config across the suite). Backed by Gemini models; security centers on Trusted Workspaces. An Orchestrator decomposes goals and spawns dynamic subagents that run async in the background.

Two products, one name: capabilities here are the IDE/CLI harness. The hosted Gemini API "Managed Agents" (ai.google.dev) is a separate surface that currently lacks custom tools and subagent delegation — do not conflate it with the IDE/CLI harness rated in this skill.

2. Config Layout

<project-root>/
├── AGENTS.md                         ← cross-tool instructions (also Cursor / Claude Code)
├── GEMINI.md                         ← Antigravity-specific (HIGHER precedence than AGENTS.md)
└── .agents/
    ├── rules/    *.md                ← Manual | Always On | Model Decision | Glob (≤12,000 chars)
    ├── skills/   <name>/SKILL.md     ← Agent Skills (Anthropic standard) [path pin via install test]
    └── hooks.json                    ← lifecycle hooks (workspace)
.antigravity/
    └── mcp.json                      ← MCP { "mcpServers": { ... } } (project-local)

~/.gemini/                            ← user-global (Gemini-namespaced)
├── GEMINI.md  ·  AGENTS.md           ← global rules (AGENTS.md cross-tool, v1.20.3+)
├── config/                           ← global hooks.json customization dir
└── antigravity/
    ├── skills/                       ← global Agent Skills
    ├── mcp_config.json               ← global MCP config
    └── global_workflows/             ← saved-prompt Workflows (slash commands, /)

Precedence: System Rules (immutable, DeepMind) > GEMINI.md > AGENTS.md. gald3r already generates AGENTS.md → baseline integration works as-is on Antigravity.

3. gald3r Integration

Cheapest high-parity install: ship gald3r's AGENTS.md + g-skl-*/SKILL.md — Antigravity reads the instruction file and discovers Anthropic-format skills natively. Put Antigravity-only must-win directives in GEMINI.md (it outranks AGENTS.md). Map @g-* commands to Workflows, g-rl-* to Always On / Model Decision rules, and wire g-hk-*.ps1 via hooks.json (before_model_call ≈ session-start, on_loop_stop ≈ stop, before_tool_call ≈ pre-tool guard). Fold g-agnt-* into rules/skills — subagents are dynamic (no file-based role discovery).

Verify

Test-Path AGENTS.md                                    # instruction file Antigravity reads
Test-Path .antigravity/mcp.json                        # project MCP (or ~/.gemini/antigravity/mcp_config.json)
Test-Path .agents/hooks.json                           # lifecycle hooks
Test-Path .agents/skills ; Test-Path .antigravity/skills ; Test-Path .agent/skills   # pin which path is live

4. Common Pitfalls

• Wrong instruction file: Antigravity reads AGENTS.md + GEMINI.md — not CLAUDE.md. For Antigravity-only overrides use GEMINI.md (outranks AGENTS.md).
• Agents don't auto-map: subagents are dynamic (Orchestrator-spawned). There is no honored .agents/agents/ dir for g-agnt-*.md — fold agent guidance into rules/skills.
• Managed Agents ≠ harness: don't rate hooks/subagents from the hosted Gemini API surface.
• Skill dir not canonical: .agents/skills vs .antigravity/skills vs .agent/skills — pin via install test before hard-coding.
• Hooks are shell scripts: official examples are .sh; wire g-hk-*.ps1 via { "type":"command", "command":"powershell -File …" }; pin payload schema via install test.
• File-size caps: rule and workflow files are limited to 12,000 chars each.

5. Capability Summary

Full assessment + evidence in PLATFORM_SPEC.md. Re-verify on the next @g-platform-scan-docs antigravity (crawl_max_age_days: 7).