[codex] migrate Gemini native CLI to Antigravity

.claude-plugin/marketplace.json

@@ -6,7 +6,7 @@
   },
   "metadata": {
     "description": "Configurable multi-model consensus code review, plan review, and general review for Claude Code and Codex",
-    "version": "1.6.0"
+    "version": "1.7.0"
   },
   "plugins": [
     {

.gitignore

@@ -8,3 +8,4 @@
 .vscode/
 node_modules/
 draft.md
+.antigravitycli/

CHANGELOG.md

@@ -4,6 +4,16 @@ All notable changes to this project will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
+## [1.7.0] - 2026-05-20
+
+### Changed
+
+- Native Google/Gemini execution now uses Antigravity CLI (`agy`) instead of the retiring Gemini CLI.
+- Claude command templates and Codex skill references route `gemini` model entries through `agy --sandbox -p`; convergence uses a fresh Antigravity prompt because `agy --continue -p` can include prior output before the new answer.
+- Codex default config now keeps the stable `gemini` model id but uses `command: "agy"` and the display name `Gemini via Antigravity`.
+- Setup docs now detect `agy` and treat bare `gemini` command values as legacy configs to migrate.
+- Plugin metadata bumped to v1.7.0.
+
 ## [1.6.0] - 2026-05-20
 
 ### Added

README.md

@@ -63,7 +63,7 @@ Alias skills are also provided for `code-review`, `plan-review`, `plan-reviwe`,
 | Claude Code or Codex | Yes | The host environment determines the lead reviewer |
 | Kilo CLI | Recommended | Routes OpenRouter models with one API key |
 | OpenRouter API key | Recommended | Required for Kilo/OpenRouter models |
-| Native CLIs | Optional | Claude supports `codex`, `gemini`, and `qwen`; Codex supports non-Codex native CLIs |
+| Native CLIs | Optional | Claude supports `codex`, Antigravity CLI (`agy`), and `qwen`; Codex supports non-Codex native CLIs |
 | CodeRabbit CLI | Optional | Supplementary static analysis for code reviews |
 
 **Minimal setup**: the host lead plus 1 external model is enough for consensus reviews.
@@ -74,7 +74,7 @@ Alias skills are also provided for `code-review`, `plan-review`, `plan-reviwe`,
 |-------|----------|---------------|---------------|----------------|
 | Claude | Anthropic | built-in | lead | not used |
 | Codex / GPT | OpenAI | `openai/gpt-5.4-codex` | `codex` | lead, not external |
-| Gemini 3.1 Pro | Google | `google/gemini-3.1-pro-preview` | `gemini` | `gemini` |
+| Gemini 3.1 Pro | Google | `google/gemini-3.1-pro-preview` | Antigravity CLI (`agy`) | Antigravity CLI (`agy`) |
 | Kimi K2.6 | Moonshot | `moonshotai/kimi-k2.6` | Kilo | Kilo |
 | Grok 4.20 | xAI | `x-ai/grok-4.20-beta` | Kilo | Kilo, disabled by default |
 | MiniMax M2.7 | MiniMax | `minimax/minimax-m2.7` | Kilo | Kilo |
@@ -98,7 +98,7 @@ Example Codex config:
 
 ```json
 {
-  "version": "1.6.0",
+  "version": "1.7.0",
   "lead": {
     "id": "codex",
     "name": "Codex",
@@ -107,8 +107,8 @@ Example Codex config:
   "models": [
     {
       "id": "gemini",
-      "name": "Gemini 3.1 Pro",
-      "command": "gemini",
+      "name": "Gemini via Antigravity",
+      "command": "agy",
       "resume_flag": "",
       "enabled": true
     },

plugins/consensus/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "consensus",
-  "version": "1.6.0",
+  "version": "1.7.0",
   "description": "Configurable multi-model code review, plan review, and general review with consensus convergence for Claude Code",
   "author": {
     "name": "AltimateAI"

plugins/consensus/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "consensus",
-  "version": "1.6.0",
+  "version": "1.7.0",
   "description": "Codex-led multi-model code review, plan review, and general review with consensus convergence",
   "author": {
     "name": "AltimateAI",

plugins/consensus/AGENTS.md

@@ -14,7 +14,7 @@ Claude commands load model configuration from:
 1. `~/.claude/consensus.json` - user config created by `/consensus-setup`
 2. `plugins/consensus/consensus.config.json` - Claude plugin defaults
 
-Claude is always the lead participant. Native CLIs (`codex`, `gemini`, `qwen`) are handled as special cases in the command templates.
+Claude is always the lead participant. Native CLIs (`codex`, Antigravity CLI via `agy`, `qwen`) are handled as special cases in the command templates. The `gemini` model id remains stable, but native Google/Gemini execution should use `agy`; bare `gemini` is legacy only.
 
 ## Codex Configuration
 

plugins/consensus/commands/code-review.md

@@ -91,12 +91,13 @@ Source the API key (targeted — only export `OPENROUTER_API_KEY`):
 For each model in `MODELS`, verify CLI availability:
 - Commands starting with `kilo` -> check: `command -v kilo` AND `[ -n "$OPENROUTER_API_KEY" ]`
 - Commands starting with `codex` -> check: `command -v codex`
-- Commands starting with `gemini` -> check: `command -v gemini`
+- Commands starting with `agy` -> check: `command -v agy`
+- Commands starting with `gemini` -> legacy config; check `command -v agy`, run through the `agy` path below, and warn the user to rerun `/consensus-setup`
 - Commands starting with `qwen` -> check: `command -v qwen`
 
 Run all checks in parallel. Remove unavailable models from `MODELS` with a warning for each:
 ```
-Warning: Skipping {model.name} — {reason: "kilo CLI not found" / "OPENROUTER_API_KEY not set" / "codex CLI not found"}
+Warning: Skipping {model.name} — {reason: "kilo CLI not found" / "OPENROUTER_API_KEY not set" / "codex CLI not found" / "Antigravity CLI not found"}
 ```
 
 **Check CodeRabbit availability:**
@@ -215,7 +216,7 @@ For each model, substitute `{MODEL_ID}`, `{MODEL_NAME}`, `{MODEL_COMMAND}`, `{MO
 
 **Building `{EXTRA_DIRS_FLAGS}`** — if `EXTRA_DIRS` is non-empty, build per-CLI flags:
 - For commands starting with `codex`: `--add-dir /path1 --add-dir /path2` (one `--add-dir` per directory)
-- For commands starting with `gemini`: `--include-directories /path1,/path2` (comma-separated)
+- For commands starting with `agy` or legacy `gemini`: `--add-dir /path1 --add-dir /path2` (one `--add-dir` per directory)
 - For commands starting with `qwen`: `--include-directories /path1,/path2` (comma-separated)
 - For commands starting with `kilo`: empty string (kilo has no flag — the paths are already in the prompt)
 
@@ -236,8 +237,8 @@ SESSION_DIR={SESSION_DIR}
    **If `{MODEL_COMMAND}` starts with `codex`:**
    codex exec -s read-only {EXTRA_DIRS_FLAGS} -o $SESSION_DIR/{MODEL_ID}.md - < $SESSION_DIR/prompt.md
 
-   **If `{MODEL_COMMAND}` starts with `gemini`:**
-   gemini {EXTRA_DIRS_FLAGS} -p "$(cat $SESSION_DIR/prompt.md)" --approval-mode plan > $SESSION_DIR/{MODEL_ID}.md 2>&1
+   **If `{MODEL_COMMAND}` starts with `agy` or legacy `gemini`:**
+   agy --sandbox {EXTRA_DIRS_FLAGS} -p "$(cat $SESSION_DIR/prompt.md)" > $SESSION_DIR/{MODEL_ID}.md 2>&1
 
    **If `{MODEL_COMMAND}` starts with `qwen`:**
    qwen {EXTRA_DIRS_FLAGS} --approval-mode plan -p "$(cat $SESSION_DIR/prompt.md)" -o text > $SESSION_DIR/{MODEL_ID}.md 2>&1
@@ -262,8 +263,8 @@ After sending the review, WAIT. The lead will send you a convergence prompt. Whe
    **If `{MODEL_COMMAND}` starts with `codex`:**
    codex exec resume --last - < $SESSION_DIR/convergence-prompt-{MODEL_ID}.md > $SESSION_DIR/{MODEL_ID}-convergence.md 2>&1
 
-   **If `{MODEL_COMMAND}` starts with `gemini`:**
-   gemini --resume latest -p "$(cat $SESSION_DIR/convergence-prompt-{MODEL_ID}.md)" --approval-mode plan > $SESSION_DIR/{MODEL_ID}-convergence.md 2>&1
+   **If `{MODEL_COMMAND}` starts with `agy` or legacy `gemini`:**
+   agy --sandbox -p "$(cat $SESSION_DIR/convergence-prompt-{MODEL_ID}.md)" > $SESSION_DIR/{MODEL_ID}-convergence.md 2>&1
 
    **If `{MODEL_COMMAND}` starts with `qwen`:**
    qwen -c -p "$(cat $SESSION_DIR/convergence-prompt-{MODEL_ID}.md)" -o text > $SESSION_DIR/{MODEL_ID}-convergence.md 2>&1
@@ -493,7 +494,7 @@ On failure: preserve `$SESSION_DIR` for debugging and tell the user where files
 8. **Dynamic quorum.** Use `MIN_QUORUM` from config. Abort if fewer than `MIN_QUORUM` reviews available (including Claude).
 9. **Convergence through messaging.** Lead sends draft to teammates, they run their model and report back. Max 2 rounds.
 10. **No plan mode.** Code reviews are presented directly, not written to plan files.
-11. **Be patient with teammates — they almost never fail.** External CLI models (Codex, Gemini, Kilo) take time to explore the codebase but almost always finish successfully. Follow this activity-based patience protocol:
+11. **Be patient with teammates — they almost never fail.** External CLI models (Codex, Antigravity, Kilo) take time to explore the codebase but almost always finish successfully. Follow this activity-based patience protocol:
     - **Poll output files** every ~1 minute using `wc -c < $SESSION_DIR/{model.id}.md 2>/dev/null || echo 0` to check file size.
     - **Growing file (or no file yet)** = the model is working. Keep waiting.
     - **A teammate is ONLY considered stuck if**: their output file exists AND its size has not changed for **10 consecutive checks** (10 minutes of zero growth).

plugins/consensus/commands/consensus-setup.md

@@ -39,7 +39,7 @@ command -v codex && echo "CODEX_OK" || echo "CODEX_MISSING"
 ```
 
 ```bash
-command -v gemini && echo "GEMINI_OK" || echo "GEMINI_MISSING"
+command -v agy && echo "AGY_OK" || echo "AGY_MISSING"
 ```
 
 ```bash
@@ -59,7 +59,7 @@ Report findings to the user:
 
 - Kilo CLI: {installed / not found}
 - Codex CLI: {installed / not found}
-- Gemini CLI: {installed / not found}
+- Antigravity CLI (`agy`): {installed / not found}
 - Qwen CLI: {installed / not found}
 - OpenRouter API key: {found in ~/.claude/.env / not found}
 ```
@@ -74,7 +74,7 @@ AskUserQuestion:
     - label: "OpenRouter (Recommended)"
       description: "1 API key, all 9 models via Kilo CLI. Simplest setup."
     - label: "Native CLIs"
-      description: "Use codex, gemini, qwen CLIs directly where available. Requires each CLI installed separately."
+      description: "Use codex, Antigravity (`agy`), and qwen CLIs directly where available. Requires each CLI installed separately."
     - label: "Both"
       description: "Mix and match — use native CLIs where available, OpenRouter/Kilo for the rest."
 ```
@@ -122,7 +122,7 @@ The 9 supported models and their CLI mappings:
 | Model ID | Name | OpenRouter (Kilo) | Native CLI |
 |----------|------|-------------------|------------|
 | `gpt` | GPT 5.4 Codex | `kilo run -m openrouter/openai/gpt-5.4-codex --auto` | `codex` (if codex CLI installed — note: native codex uses whatever model your `~/.codex/config.toml` specifies, e.g. `gpt-5.5`) |
-| `gemini` | Gemini 3.1 Pro | `kilo run -m openrouter/google/gemini-3.1-pro-preview --auto` | `gemini` (if gemini CLI installed) |
+| `gemini` | Gemini via Antigravity | `kilo run -m openrouter/google/gemini-3.1-pro-preview --auto` | `agy` (if Antigravity CLI installed) |
 | `kimi` | Kimi K2.6 | `kilo run -m openrouter/moonshotai/kimi-k2.6 --auto` | OpenRouter only |
 | `grok` | Grok 4.20 | `kilo run -m openrouter/x-ai/grok-4.20-beta --auto` | OpenRouter only |
 | `minimax` | MiniMax M2.7 | `kilo run -m openrouter/minimax/minimax-m2.7 --auto` | OpenRouter only |
@@ -131,11 +131,11 @@ The 9 supported models and their CLI mappings:
 | `mimo` | MiMo V2 Pro | `kilo run -m openrouter/xiaomi/mimo-v2-pro --auto` | OpenRouter only |
 | `deepseek` | DeepSeek V4 Pro | `kilo run -m openrouter/deepseek/deepseek-v4-pro --auto` | OpenRouter only |
 
-**Note on native CLIs**: For `codex`, `gemini`, and `qwen`, set the config's `command` field to just `codex`, `gemini`, or `qwen`. The teammate template in the review/plan commands detects these and uses the correct native invocation patterns automatically (e.g., `codex exec -s read-only` for reviews, `codex exec resume --last` for convergence, `gemini -p` for reviews, `gemini --resume latest` for convergence, `qwen --approval-mode plan -p` with `-o text` for reviews, `qwen -c -p` for convergence). The `resume_flag` field is ignored for native CLIs.
+**Note on native CLIs**: For `codex`, `agy`, and `qwen`, set the config's `command` field to just `codex`, `agy`, or `qwen`. The teammate template in the review/plan commands detects these and uses the correct native invocation patterns automatically (e.g., `codex exec -s read-only` for reviews, `codex exec resume --last` for convergence, `agy --sandbox -p` for reviews and convergence prompts, `qwen --approval-mode plan -p` with `-o text` for reviews, `qwen -c -p` for convergence). The `resume_flag` field is ignored for native CLIs. If an existing config still contains `command: "gemini"`, treat it as legacy and migrate it to `agy` during reconfiguration.
 
 Determine which models are available:
 - **OpenRouter path**: All 9 available if `kilo` installed + API key set
-- **Native path**: Only `gpt` (if codex installed), `gemini` (if gemini installed), and `qwen` (if qwen installed)
+- **Native path**: Only `gpt` (if codex installed), `gemini` (if `agy` installed), and `qwen` (if qwen installed)
 - **Both path**: Native CLI where available, OpenRouter/Kilo for the rest
 
 ```
@@ -147,7 +147,7 @@ AskUserQuestion:
     - label: "GPT 5.4 Codex"
       description: "{available via OpenRouter / available via codex CLI / not available}"
     - label: "Gemini 3.1 Pro"
-      description: "{available via OpenRouter / available via gemini CLI / not available}"
+      description: "{available via OpenRouter / available via Antigravity CLI / not available}"
     - label: "Kimi K2.6"
       description: "{available via OpenRouter / not available}"
     - label: "Grok 4.20"
@@ -208,7 +208,7 @@ Write the config to `~/.claude/consensus.json` using the Write tool:
 
 ```json
 {
-  "version": "1.6.0",
+  "version": "1.7.0",
   "models": [
     {
       "id": "{model_id}",
@@ -237,9 +237,15 @@ For each enabled model, run a quick test:
 
 ```bash
 [ -f ~/.claude/.env ] && export OPENROUTER_API_KEY=$(grep '^OPENROUTER_API_KEY=' ~/.claude/.env | cut -d= -f2- | tr -d '"')
-{model.command} "Reply with exactly: PONG" 2>&1 | head -20
 ```
 
+Use the command pattern for the configured CLI:
+
+- `codex`: `printf '%s\n' "Reply with exactly: PONG" | codex exec -s read-only - 2>&1 | head -20`
+- `agy` or legacy `gemini`: `agy --sandbox -p "Reply with exactly: PONG" 2>&1 | head -20`
+- `qwen`: `qwen --approval-mode plan -p "Reply with exactly: PONG" -o text 2>&1 | head -20`
+- Kilo/OpenRouter: `{model.command} "Reply with exactly: PONG" 2>&1 | head -20`
+
 Check if the output contains "PONG" (case-insensitive).
 
 - **Pass**: Report success for that model
@@ -292,7 +298,7 @@ Print the final summary:
 
 1. **9 fixed models only.** Do not offer custom model configuration. The wizard supports exactly the 9 models listed above.
 2. **Idempotent .env updates.** When writing API keys, preserve all existing keys in the file. Only add/update the `OPENROUTER_API_KEY` line.
-3. **Native CLI support for codex, gemini, and qwen.** When a user selects native CLIs, set the config's `command` field to `codex`, `gemini`, or `qwen`. The teammate template in the review/plan commands handles the full invocation patterns automatically. All other models use OpenRouter/Kilo only.
+3. **Native CLI support for codex, agy, and qwen.** When a user selects native CLIs, set the config's `command` field to `codex`, `agy`, or `qwen`. The teammate template in the review/plan commands handles the full invocation patterns automatically. All other models use OpenRouter/Kilo only. Bare `gemini` is a legacy config value; migrate it to `agy` when rewriting the config.
 4. **OpenRouter is the recommended path.** 1 key = 9 models. Emphasize this as the simplest setup.
 5. **Enforce minimum 1 external model.** Claude alone is not a consensus.
 6. **Hard-stop on quorum failure.** Never finalize a config that can't meet its own quorum.

plugins/consensus/commands/plan-review.md

@@ -91,12 +91,13 @@ Source the API key (targeted — only export `OPENROUTER_API_KEY`):
 For each model in `MODELS`, verify CLI availability:
 - Commands starting with `kilo` -> check: `command -v kilo` AND `[ -n "$OPENROUTER_API_KEY" ]`
 - Commands starting with `codex` -> check: `command -v codex`
-- Commands starting with `gemini` -> check: `command -v gemini`
+- Commands starting with `agy` -> check: `command -v agy`
+- Commands starting with `gemini` -> legacy config; check `command -v agy`, run through the `agy` path below, and warn the user to rerun `/consensus-setup`
 - Commands starting with `qwen` -> check: `command -v qwen`
 
 Run all checks in parallel. Remove unavailable models from `MODELS` with a warning for each:
 ```
-Warning: Skipping {model.name} — {reason: "kilo CLI not found" / "OPENROUTER_API_KEY not set" / "codex CLI not found"}
+Warning: Skipping {model.name} — {reason: "kilo CLI not found" / "OPENROUTER_API_KEY not set" / "codex CLI not found" / "Antigravity CLI not found"}
 ```
 
 Count available models + 1 (Claude) = `TOTAL_PARTICIPANTS`.
@@ -228,7 +229,7 @@ For each model, substitute `{MODEL_ID}`, `{MODEL_NAME}`, `{MODEL_COMMAND}`, `{MO
 
 **Building `{EXTRA_DIRS_FLAGS}`** — if `EXTRA_DIRS` is non-empty, build per-CLI flags:
 - For commands starting with `codex`: `--add-dir /path1 --add-dir /path2` (one `--add-dir` per directory)
-- For commands starting with `gemini`: `--include-directories /path1,/path2` (comma-separated)
+- For commands starting with `agy` or legacy `gemini`: `--add-dir /path1 --add-dir /path2` (one `--add-dir` per directory)
 - For commands starting with `qwen`: `--include-directories /path1,/path2` (comma-separated)
 - For commands starting with `kilo`: empty string (kilo has no flag — the paths are already in the prompt)
 
@@ -249,8 +250,8 @@ SESSION_DIR={SESSION_DIR}
    **If `{MODEL_COMMAND}` starts with `codex`:**
    codex exec -s read-only {EXTRA_DIRS_FLAGS} -o $SESSION_DIR/{MODEL_ID}.md - < $SESSION_DIR/prompt.md
 
-   **If `{MODEL_COMMAND}` starts with `gemini`:**
-   gemini {EXTRA_DIRS_FLAGS} -p "$(cat $SESSION_DIR/prompt.md)" --approval-mode plan > $SESSION_DIR/{MODEL_ID}.md 2>&1
+   **If `{MODEL_COMMAND}` starts with `agy` or legacy `gemini`:**
+   agy --sandbox {EXTRA_DIRS_FLAGS} -p "$(cat $SESSION_DIR/prompt.md)" > $SESSION_DIR/{MODEL_ID}.md 2>&1
 
    **If `{MODEL_COMMAND}` starts with `qwen`:**
    qwen {EXTRA_DIRS_FLAGS} --approval-mode plan -p "$(cat $SESSION_DIR/prompt.md)" -o text > $SESSION_DIR/{MODEL_ID}.md 2>&1
@@ -275,8 +276,8 @@ After sending the plan, WAIT. The lead will send you a convergence prompt. When
    **If `{MODEL_COMMAND}` starts with `codex`:**
    codex exec resume --last - < $SESSION_DIR/convergence-prompt-{MODEL_ID}.md > $SESSION_DIR/{MODEL_ID}-convergence.md 2>&1
 
-   **If `{MODEL_COMMAND}` starts with `gemini`:**
-   gemini --resume latest -p "$(cat $SESSION_DIR/convergence-prompt-{MODEL_ID}.md)" --approval-mode plan > $SESSION_DIR/{MODEL_ID}-convergence.md 2>&1
+   **If `{MODEL_COMMAND}` starts with `agy` or legacy `gemini`:**
+   agy --sandbox -p "$(cat $SESSION_DIR/convergence-prompt-{MODEL_ID}.md)" > $SESSION_DIR/{MODEL_ID}-convergence.md 2>&1
 
    **If `{MODEL_COMMAND}` starts with `qwen`:**
    qwen -c -p "$(cat $SESSION_DIR/convergence-prompt-{MODEL_ID}.md)" -o text > $SESSION_DIR/{MODEL_ID}-convergence.md 2>&1
@@ -476,7 +477,7 @@ On failure: preserve `$SESSION_DIR` for debugging and tell the user where files
 8. **Dynamic quorum.** Use `MIN_QUORUM` from config. Abort if fewer than `MIN_QUORUM` plans available (including Claude).
 9. **Return to plan mode.** After final plan + cleanup, call `EnterPlanMode` for user review (with fallback to direct presentation).
 10. **Convergence through messaging.** Lead sends draft to teammates, they run their model and report back. Max 2 rounds.
-11. **Be patient with teammates — they almost never fail.** External CLI models (Codex, Gemini, Kilo) take time to explore the codebase but almost always finish successfully. Follow this activity-based patience protocol:
+11. **Be patient with teammates — they almost never fail.** External CLI models (Codex, Antigravity, Kilo) take time to explore the codebase but almost always finish successfully. Follow this activity-based patience protocol:
     - **Poll output files** every ~1 minute using `wc -c < $SESSION_DIR/{model.id}.md 2>/dev/null || echo 0` to check file size.
     - **Growing file (or no file yet)** = the model is working. Keep waiting.
     - **A teammate is ONLY considered stuck if**: their output file exists AND its size has not changed for **10 consecutive checks** (10 minutes of zero growth).