feat: add Antigravity CLI harness with Google OAuth

AGENTS.md

@@ -6,7 +6,7 @@ Instructions for coding agents working on qmax-code.
 
 qmax-code is a terminal **QA** agent that can also host coding CLIs. It runs
 standalone on a local repository or connected to QualityMax. Through `/orch`
-it can launch Claude Code, Codex, or OpenCode while keeping qmax tools and
+it can launch Claude Code, Codex, Antigravity, or OpenCode while keeping qmax tools and
 terminal UX — or run the built-in loop on Anthropic, Cerebras, or Ollama.
 
 This repository is **Go 1.25.13+**, not TypeScript. The thing users install is one
@@ -23,7 +23,8 @@ those other agents when the job is coding.
 - **Cerebras** = fast inference (~1000–2000+ tok/s). Models today: GPT-OSS
   120B, GLM 4.7, Gemma 4. **Qwen 3.8 is coming soon** on Cerebras; do not
   wire it into `/orch` until the model ID is live on the Cerebras API.
-- **Claude Code / Codex** = judgment for hard design and tricky refactors.
+- **Claude Code / Codex / Antigravity** = judgment for hard design and tricky refactors.
+  Antigravity uses Google OAuth (run interactive `agy` to sign in), not a Gemini API key.
 - **OpenCode** = opt-in providers (Z.AI GLM, Groq, OpenRouter) via
   `/providers`.
 

CHANGELOG.md

@@ -4,6 +4,14 @@ All notable changes to qmax-code. Versions follow [Semantic Versioning](https://
 
 ## [Unreleased]
 
+### Added
+- Antigravity CLI (`agy`) is a first-class `/orch` backend. It runs Google's
+  native agent harness as a subprocess (`agy -p --output-format stream-json`),
+  attaches qmax tools through `~/.gemini/config/mcp_config.json`, and authenticates
+  with **Google OAuth** (interactive `agy` browser sign-in) using the same Google account as
+  AI Studio. A Gemini API key is not required. Select it with `/agy`,
+  `/orch`, or `--backend agy`.
+
 ## [1.32.0] - 2026-09-05
 
 ### Added

CLAUDE.md

@@ -2,11 +2,11 @@
 
 qmax-code is a **Go** terminal **QA** agent (Go 1.25.13+). It ships as one
 compiled binary: `curl | bash` installs a file, not a Node or Python runtime.
-Claude Code, Codex, and OpenCode are separate subprocesses that `/orch` can
+Claude Code, Codex, Antigravity, and OpenCode are separate subprocesses that `/orch` can
 host; they are not this repository's runtime.
 
 **Positioning:** not an IDE, not a faster Claude. Cursor owns the editor.
-Claude Code / Codex own hard coding. qmax-code owns QA (tests, crawls, review,
+Claude Code / Codex / Antigravity own hard coding. qmax-code owns QA (tests, crawls, review,
 receipts) and switches inference in `/orch`. **Cerebras** is the fast path
 (~1000–2000+ tok/s: GPT-OSS 120B, GLM 4.7, Gemma 4; **Qwen 3.8 coming soon**).
 Do not add Qwen to the picker until Cerebras publishes the live model ID.

CONTRIBUTING.md

@@ -27,13 +27,13 @@ Thanks for your interest in improving `qmax-code`. This document covers everythi
 | Requirement | Notes |
 |---|---|
 | **Go 1.25.13+** | See `go.mod` for the exact version. `go version` to check. |
-| **Inference backend** | The direct API path needs an Anthropic key. You can instead develop against a logged-in Claude Code/Codex CLI, an enabled OpenCode provider, Cerebras, or Ollama. |
+| **Inference backend** | The direct API path needs an Anthropic key. You can instead develop against a logged-in Claude Code/Codex/Antigravity CLI, an enabled OpenCode provider, Cerebras, or Ollama. |
 | **QualityMax account** | Required only for connected cloud tools (test generation, crawl, repo review). Standalone `--local` development and most unit tests do not need one. |
 
 Optional but useful:
 
 - [golangci-lint](https://golangci-lint.run/usage/install/) — the CI linter; run it locally to catch issues before pushing
-- [Claude Code](https://claude.ai/download), [Codex](https://github.com/openai/codex), or [OpenCode](https://opencode.ai) — only needed when working on that CLI orchestration backend
+- [Claude Code](https://claude.ai/download), [Codex](https://github.com/openai/codex), [Antigravity](https://antigravity.google/docs/cli/install/), or [OpenCode](https://opencode.ai) — only needed when working on that CLI orchestration backend
 - [Ollama](https://ollama.com) or a Cerebras key — only needed when working on that inference adapter
 
 ---

README.md

@@ -33,7 +33,7 @@ repositories, heal scripts, and prepare CI. It calls the QualityMax API
 directly, so the separate `qmax` CLI is optional.
 
 Use the built-in agent with Anthropic, Cerebras, or Ollama, or use
-**orchestration mode** to run Claude Code, Codex, or OpenCode with the same qmax
+**orchestration mode** to run Claude Code, Codex, Antigravity, or OpenCode with the same qmax
 QA tools through MCP.
 
 ![Five open QualityMax tools and where qmax-code sits among them](docs/img/family.svg)
@@ -175,7 +175,7 @@ background-job health—remain experimental and are only exposed when
   multimodal input for Gemma 4, optional reasoning effort, and live speed
   metrics.
 - **27 managed QA skills:** the catalog is refreshed into Claude Code, Codex,
-  and OpenCode and can be inspected or reinstalled with `/skills`.
+  OpenCode, and Antigravity and can be inspected or reinstalled with `/skills`.
 - **Improved terminal sessions:** a stable input panel, prompt queue, session
   status and cost metrics, compact/verbose output toggle, ten themes, saved
   sessions, and optional cloud sync.
@@ -336,6 +336,7 @@ troubleshooting.
 | Anthropic API | `/api` or `/orch` | `ANTHROPIC_API_KEY` or OS keychain | Built-in agent loop; tool set follows connected vs. standalone mode. |
 | Claude Code | `/cc` or `/orch` | Local Claude Code login | CLI subprocess; qmax tools arrive through MCP. Uses Claude Code authentication and plan/provider billing; see [subscription setup](docs/ORCHESTRATION.md#claude-code-subscription-billing). |
 | Codex | `/codex` or `/orch` | Local Codex login | CLI subprocess using the user's OpenAI access; qmax tools arrive through MCP. |
+| Antigravity | `/agy` or `/orch` | Google OAuth (run `agy` once to sign in) | CLI subprocess using the same Google account as AI Studio. No Gemini API key. qmax tools arrive through MCP. |
 | Cerebras | `/gemma`, `/orch`, or `--backend cerebras` | `CEREBRAS_API_KEY` or OS keychain | Built-in native function calling. Fast inference (~1000–2000+ tok/s): GPT-OSS 120B, GLM 4.7, Gemma 4 (vision + effort). **Qwen 3.8 coming soon.** |
 | OpenCode | `/opencode` or `/orch` | Per-provider key in OS keychain | CLI subprocess for opt-in Z.AI, Groq, and OpenRouter providers. |
 | Ollama | `/ollama` or `/orch` | Configured Ollama endpoint | Self-hosted inference; configure the URL and model first. |
@@ -373,6 +374,7 @@ agent. qmax-code declares that dependency in the Codex skill metadata.
 
 ```text
 /orch                    Pick backend, model, and effort
+/agy                     Switch to Antigravity (Google OAuth)
 /providers               List opt-in OpenCode providers
 /providers enable groq   Store a provider key and enable its models
 /skills                   Show managed QA skills and install status

SECURITY.md

@@ -51,10 +51,13 @@ qmax tools supplied through MCP.
 Only use Unattended mode in a trusted, recoverable workspace. Neither mode
 creates a sandbox or worktree boundary.
 
-Claude Code and Codex can optionally receive a user-level qmax MCP entry in
-`~/.claude/settings.json` or `~/.codex/config.toml`. That makes qmax tools
-available whenever the CLI is launched, not only inside qmax-code. OpenCode
-uses a separate qmax-managed overlay at `~/.qmax-code/opencode.json`.
+Claude Code, Codex, and Antigravity can optionally receive a user-level qmax MCP
+entry in `~/.claude/settings.json`, `~/.codex/config.toml`, or
+`~/.gemini/config/mcp_config.json`. That makes qmax tools available whenever
+the CLI is launched, not only inside qmax-code. OpenCode uses a separate
+qmax-managed overlay at `~/.qmax-code/opencode.json`. Antigravity authenticates
+with Google OAuth via interactive `agy` (browser sign-in); qmax-code does not store a Gemini API
+key for that backend.
 
 qmax-code also installs managed QA skills into the selected CLI's user-level
 skills directory. See [Orchestration mode](docs/ORCHESTRATION.md) for the

config_command.go

@@ -174,6 +174,12 @@ func printConfig() {
 		} else {
 			fmt.Print("  (WARNING: opencode binary not found in PATH)")
 		}
+	case "agy":
+		if bin := agent.FindAgy(); bin != "" {
+			fmt.Printf("  (agy found: %s; Google OAuth)", bin)
+		} else {
+			fmt.Print("  (WARNING: agy binary not found in PATH)")
+		}
 	}
 	fmt.Println()
 	if len(cfg.EnabledProviders) > 0 {
@@ -276,10 +282,10 @@ func setConfigField(key, value string) error {
 
 	case "backend":
 		switch value {
-		case "", "api", "cc", "codex", "cerebras", "opencode":
+		case "", "api", "cc", "codex", "agy", "cerebras", "opencode":
 			cfg.Backend = value
 		default:
-			return fmt.Errorf("invalid backend %q; allowed: api, cc, codex, cerebras, opencode", value)
+			return fmt.Errorf("invalid backend %q; allowed: api, cc, codex, agy, cerebras, opencode", value)
 		}
 
 	case "cerebras_key":

config_command_test.go

@@ -45,7 +45,7 @@ func TestSetConfigField_DefaultFrameworkRejectsBadValues(t *testing.T) {
 func TestSetConfigField_BackendValidation(t *testing.T) {
 	withTempHome(t)
 
-	for _, b := range []string{"api", "cc", "codex", "cerebras", ""} {
+	for _, b := range []string{"api", "cc", "codex", "agy", "cerebras", ""} {
 		if err := setConfigField("backend", b); err != nil {
 			t.Errorf("backend %q should be accepted: %v", b, err)
 		}

docs/COMMANDS.md

@@ -34,7 +34,7 @@ qmax-code serve --mcp
 | `--verbose` | Show tool calls and raw responses. |
 | `--professional` | Disable the cat personality for this run. |
 | `-q` | Reserved for a future quiet/CI output mode; currently has no effect. |
-| `--backend NAME` | Override the saved backend with `api`, `cc`, `codex`, `cerebras`, or `opencode`. |
+| `--backend NAME` | Override the saved backend with `api`, `cc`, `codex`, `agy`, `cerebras`, or `opencode`. |
 | `--version` | Print the version and exit. |
 
 Non-interactive stdin must include either `-p` or a positional prompt. This
@@ -87,7 +87,7 @@ qmax-code codex connect
 
 These attach the active local Claude Code or a fresh Codex login to the
 currently authenticated QualityMax user. They are separate from selecting a
-backend with `/cc` or `/codex`.
+backend with `/cc`, `/codex`, or `/agy`.
 
 ## Configuration subcommand
 
@@ -151,6 +151,7 @@ prints the selected local manifest. `verify` checks its signature offline.
 | `/api` | Switch to the direct Anthropic API. |
 | `/cc` | Switch to the Claude Code CLI backend. |
 | `/codex` | Switch to the Codex CLI backend. |
+| `/agy` | Switch to the Antigravity CLI backend (Google OAuth). |
 | `/opencode` | Switch to the OpenCode CLI backend. |
 | `/gemma [none\|low\|medium\|high]` | Activate Gemma 4 on Cerebras with the chosen reasoning level. |
 | `/gemma off` | Return to the direct Anthropic API. |
@@ -174,7 +175,7 @@ behavior.
 | `/status` | Show connection, session, usage, and model information. |
 | `/cost` | Show token usage and estimated model cost. |
 | `/update` | Check for a newer qmax-code release and install it when found. |
-| `/plan` | Show the subscription coding-plan usage window: elapsed, turns, and time until the rolling 5-hour limit resets (cc/codex/opencode backends). |
+| `/plan` | Show the subscription coding-plan usage window: elapsed, turns, and time until the rolling 5-hour limit resets (cc/codex/agy/opencode backends). |
 
 In standalone mode, `/connect` and `/project` explain how to return to connected
 mode. `/status`, `/context`, and `/config` identify the active standalone
@@ -200,7 +201,7 @@ after the current turn.
 | Command | Action |
 | --- | --- |
 | `/skills` | List all qmax QA skills and their install status by CLI backend. |
-| `/skills install` | Refresh skills for Claude Code, Codex, and OpenCode. |
+| `/skills install` | Refresh skills for Claude Code, Codex, OpenCode, and Antigravity. |
 | `/config` | Show selected session configuration. |
 | `/set KEY VALUE` | Change a supported setting from the REPL. |
 | `/keys` | Open the interactive API-key menu. |