feat: model selection for droid, agy, cursor and trae

CHANGELOG.md

@@ -2,13 +2,15 @@
 
 ## Unreleased
 
+- The `agy` install hint now points at the official `https://antigravity.google/cli/install.sh`;
+  the npm package it named does not exist.
 - Configure the model each agent runs with: `jury agents model <agent> <model>` saves a default in
   `~/.jury/config.json`, a `model` field per agent in `jury.config.json` sets one per repository,
   and `--model <agent>=<model>` sets one for a run, each overriding the one before. The model is
   passed as a flag or environment variable only when Jury runs the agent, on first runs, resumed
   rounds, replies and judging alike; the agent's own configuration is never changed. Supported for
-  codex, claude, grok, opencode, qwen, copilot and kimi; a configured model for any other agent
-  stops the run with an error. Runs record the models used, and `jury agents` shows them (#99).
+  every built-in agent except amp, which has no per-run model flag; a configured model for amp
+  stops the run with an error. Runs record the models used, and `jury agents` shows them (#99, #103, #104, #108, #111).
 - Add `jury agents jury` to save default reviewers globally in `~/.jury/config.json`, beside the
   global judge: `jury agents jury claude,droid`, `--reset`, and a checkbox picker when run with no
   arguments in a terminal. Runs without `--jury` use them; repository `reviewer` roles and

README.md

@@ -197,7 +197,7 @@ See [models](docs/configuration.md#models) for each agent's flag and the precede
 | `claude` | Anthropic Claude Code | `npm i -g @anthropic-ai/claude-code` | reviewer |
 | `grok` | xAI Grok | `npm i -g @vibe-kit/grok-cli` | reviewer |
 | `droid` | Factory Droid | `curl -fsSL https://app.factory.ai/cli \| sh` | reviewer |
-| `agy` | Google Antigravity | see Antigravity's docs | reviewer |
+| `agy` | Google Antigravity | `curl -fsSL https://antigravity.google/cli/install.sh \| bash` | reviewer |
 | `qwen` | Qwen Code | `npm i -g @qwen-code/qwen-code` | opt-in |
 | `copilot` | GitHub Copilot CLI | `npm i -g @github/copilot` | opt-in |
 | `opencode` | OpenCode | `npm i -g opencode-ai` | opt-in |

docs/configuration.md

@@ -91,8 +91,11 @@ one stops the run before any agent starts rather than silently using its default
 | `qwen` | `--model` | qwen 0.24.4 |
 | `copilot` | `--model` | Copilot CLI 1.0.88 |
 | `kimi` | `--model`, an alias defined in `~/.kimi-code/config.toml` | Kimi Code 2.1.1 |
+| `droid` | `--model` after `exec` (`droid exec --help` lists the IDs) | droid 0.226.2 |
+| `agy` | `--model` (`agy models` lists them) | agy 1.2.10 |
+| `cursor` | `--model` (`cursor-agent --list-models` lists them) | cursor-agent 2026.09.23 |
+| `trae` | `--model` after `exec` | traecli 0.207.1 |
 | `amp` | not supported: `--mode` selects model, prompt and tools together | — |
-| `droid`, `agy`, `cursor`, `trae` | not yet verified; set the model in the CLI's own configuration | — |
 
 A built-in definition supports models by marking the flag's position with a `"{{modelArgs}}"`
 element in every command it runs and giving `modelArgs` (for example `["--model", "{{model}}"]`),

lib/agents/agy.json

@@ -6,6 +6,7 @@
   "cwd": "worktree",
   "argv": [
     "agy",
+    "{{modelArgs}}",
     "--dangerously-skip-permissions",
     "--add-dir",
     "{{worktree}}",
@@ -22,6 +23,7 @@
     "supported": true,
     "argv": [
       "agy",
+      "{{modelArgs}}",
       "--dangerously-skip-permissions",
       "--add-dir",
       "{{worktree}}",
@@ -36,9 +38,12 @@
     ],
     "idFrom": "\"conversation_id\"\\s*:\\s*\"([^\"]+)\""
   },
-  "modelNote": "no per-run model flag verified for this CLI yet; set its model in the CLI's own configuration",
+  "modelArgs": [
+    "--model",
+    "{{model}}"
+  ],
   "report": "agy-json",
   "expectSeconds": 420,
-  "install": "npm install -g @google/antigravity-cli",
+  "install": "curl -fsSL https://antigravity.google/cli/install.sh | bash",
   "docs": "https://antigravity.google/docs/cli"
 }

lib/agents/cursor.json

@@ -7,6 +7,7 @@
   "cwd": "worktree",
   "argv": [
     "cursor-agent",
+    "{{modelArgs}}",
     "-p",
     "--output-format",
     "json",
@@ -15,6 +16,7 @@
   ],
   "judgeArgv": [
     "cursor-agent",
+    "{{modelArgs}}",
     "-p",
     "--output-format",
     "json",
@@ -28,7 +30,10 @@
     "supported": false,
     "reason": "fresh conversation with finding context; never extract session ids from assistant prose"
   },
-  "modelNote": "no per-run model flag verified for this CLI yet; set its model in the CLI's own configuration",
+  "modelArgs": [
+    "--model",
+    "{{model}}"
+  ],
   "report": "result-json",
   "expectSeconds": 400,
   "install": "curl https://cursor.com/install -fsS | bash",

lib/agents/droid.json

@@ -7,6 +7,7 @@
   "argv": [
     "droid",
     "exec",
+    "{{modelArgs}}",
     "--cwd",
     "{{worktree}}",
     "--auto",
@@ -23,6 +24,7 @@
     "argv": [
       "droid",
       "exec",
+      "{{modelArgs}}",
       "--cwd",
       "{{worktree}}",
       "--auto",
@@ -36,7 +38,10 @@
     ],
     "idFrom": "\"session_id\"\\s*:\\s*\"([^\"]+)\""
   },
-  "modelNote": "no per-run model flag verified for this CLI yet; set its model in the CLI's own configuration",
+  "modelArgs": [
+    "--model",
+    "{{model}}"
+  ],
   "report": "result-json",
   "expectSeconds": 130,
   "install": "curl -fsSL https://app.factory.ai/cli | sh",

lib/agents/trae.json

@@ -8,6 +8,7 @@
   "argv": [
     "traecli",
     "exec",
+    "{{modelArgs}}",
     "--sandbox",
     "read-only",
     "--skip-git-repo-check",
@@ -17,6 +18,7 @@
   "judgeArgv": [
     "traecli",
     "exec",
+    "{{modelArgs}}",
     "--sandbox",
     "workspace-write",
     "--skip-git-repo-check",
@@ -29,7 +31,10 @@
     "supported": false,
     "reason": "exec session ids are not documented in its output; replies start fresh with finding context"
   },
-  "modelNote": "no per-run model flag verified for this CLI yet; set its model in the CLI's own configuration",
+  "modelArgs": [
+    "--model",
+    "{{model}}"
+  ],
   "report": "whole",
   "expectSeconds": 900,
   "install": "sh -c \"$(curl -fsSL https://trae.cn/trae-cli/install_v2.sh)\"",

test/models.test.js

@@ -23,14 +23,20 @@ async function scratch(t, prefix) {
 // Where each CLI's model flag lands, verified against its own --help:
 // codex-cli 0.156.1 (exec and exec resume: -m, --model), Claude Code 2.1.281
 // (--model), opencode 1.18.32 (run: -m, --model), qwen 0.24.4 (-m, --model),
-// Copilot CLI 1.0.88 (--model), Kimi Code 2.1.1 (-m, --model).
+// Copilot CLI 1.0.88 (--model), Kimi Code 2.1.1 (-m, --model), droid 0.226.2
+// (exec: -m, --model), agy 1.2.10 (--model), cursor-agent 2026.09.23 (--model),
+// traecli 0.207.1 (exec: -m, --model).
 const FLAG_AT = {
   codex: { argv: 2, judgeArgv: 2, resume: 3 },
   claude: { argv: 1, judgeArgv: 1, resume: 1 },
   opencode: { argv: 2, judgeArgv: 2, resume: 2 },
   qwen: { argv: 1, judgeArgv: 1, resume: 1 },
   copilot: { argv: 1, judgeArgv: 1 },
   kimi: { argv: 1, judgeArgv: 1, resume: 1 },
+  droid: { argv: 2, resume: 2 },
+  agy: { argv: 1, resume: 1 },
+  cursor: { argv: 1, judgeArgv: 1 },
+  trae: { argv: 2, judgeArgv: 2 },
 };
 const byName = new Map(DEFAULTS.agents.map(a => [a.name, a]));
 
@@ -54,8 +60,8 @@ test("each supported agent gets its model flag in every command, and nothing wit
   assert.deepEqual(withModel(grok.argv, { ...grok, model: "grok-4.3" }), grok.argv);
 });
 
-test("agents without a verified per-run model say so and refuse a configured model", () => {
-  for (const name of ["amp", "droid", "agy", "cursor", "trae"]) {
+test("agents without a per-run model say so and refuse a configured model", () => {
+  for (const name of ["amp"]) {
     const agent = byName.get(name);
     assert.equal(supportsModel(agent), false, name);
     assert.ok(agent.modelNote, `${name} explains why`);