makaron-cli 0.13.6 → 0.13.7

package/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "makaron-cli",
-  "version": "0.13.6",
+  "version": "0.13.7",
   "description": "Give Claude Code a creative agent. Pass complete creative requests and source media to Makaron Chat.",
   "author": {
     "name": "Versa AI",

package/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "makaron-cli",
-  "version": "0.13.6",
+  "version": "0.13.7",
   "description": "Give Codex a creative agent. Pass complete creative requests and source media to Makaron Chat.",
   "author": {
     "name": "Versa AI",

package/README.md

@@ -98,7 +98,12 @@ npx makaron-cli chat --project auto --image photo.jpg --json -b "make it cinemat
 npx makaron-cli chat --project auto --image img1.jpg --image img2.jpg --json -b "combine these"
 ```
 
-`chat` always routes agent, image, and video models automatically. Never pass `--agent-model`, `--image-model`, `--video-model`, or the legacy `--model` flag to `chat`; the CLI rejects them before starting a run. Model flags remain available only on explicit low-level commands such as `edit` and `video create`.
+`chat` routes image and video models automatically, but you may select the Agent LLM with `--agent-model`. Accepted values are exactly `auto`, `gpt-5.6-terra`, `gpt-5.6-sol`, `gpt-5.6-luna`, `grok-4.5`, and `deepseek-v4-pro`; `auto` currently resolves to `gpt-5.6-terra`. This flag changes only the reasoning/tool-calling Agent LLM. The CLI still rejects `--image-model`, `--video-model`, and the legacy `--model` flag before starting a chat run, so image/video model names cannot be mistaken for Agent models.
+
+```bash
+# Explicit lower-cost Agent LLM for a controlled comparison
+npx makaron-cli chat --project auto --agent-model deepseek-v4-pro --json -b "make a 20s badminton video"
+```
 
 Returns immediately:
 ```json
@@ -342,7 +347,7 @@ npx makaron-cli video create --script "Shot 1 (15s): <<<image_1>>> and <<<image_
 
 # 3b. Native SeeDance or MiniMax H3 text-to-video (no image required)
 npx makaron-cli video create --script "Shot 1 (5s): A neon one-person studio wakes at dawn" --duration 5 --video-model seedance-fast --aspect 16:9
-npx makaron-cli video create --script "Shot 1 (15s): A premium creative editor comes alive" --duration 15 --video-model minimax-h3 --video-resolution 2k --aspect 16:9
+npx makaron-cli video create --script "Shot 1 (15s): A premium creative editor comes alive" --duration 15 --video-model minimax-h3 --aspect 16:9
 
 # 3c. Edit a video from a local file or public URL
 npx makaron-cli video create --script "make it funny" --video input.mp4 --duration 5 --video-model seedance-fast
@@ -358,7 +363,7 @@ For project/timeline video editing, use:
 npx makaron-cli chat --project <id|auto> --video input.mp4 -b "make it funny"
 ```
 
-Options for `video create`: `--script "..."`, `--script-file <path>`, `--image <url>` (repeatable, up to the selected model limit), `--video <file|url>` and `--audio <file|url>` (repeatable where supported), `--duration <seconds>`, `--aspect 9:16|16:9|1:1`, `--video-model seedance-fast|seedance-mini|seedance|seedance-2.5|kling|grok|google-omni|minimax-h3`, `--video-resolution auto|480p|720p|768p|1080p|2k|4k`. Default model is `seedance-fast`. SeeDance accepts native text-to-video with no image and integer output duration 4-15s (default 5s); `seedance-mini` supports 480p/720p and is best for cheaper drafts/multi-size tests; MiniMax H3 accepts native text-to-video, 4-15s output, and up to 9 image, up to 3 video, and up to 3 audio references through Makaron Agent/chat. H3 defaults to public 2k output; 768p requires provider preview access plus `MINIMAX_H3_ENABLE_768P=true` on the Makaron server. Kling supports 5-15s; Grok 1.5 supports 1-15s single-image-to-video only; Gemini Omni supports 3-10s fast 720p image/video generation and editing with native generated audio, including up to 6 image references when no video reference is provided. For `--video-model grok`, forced `--aspect` is ignored to avoid xAI stretching the source image; pad/create the image at the target shape first or use another model.
+Options for `video create`: `--script "..."`, `--script-file <path>`, `--image <url>` (repeatable, up to the selected model limit), `--video <file|url>` and `--audio <file|url>` (repeatable where supported), `--duration <seconds>`, `--aspect 9:16|16:9|1:1`, `--video-model seedance-fast|seedance-mini|seedance|seedance-2.5|kling|grok|google-omni|minimax-h3`, `--video-resolution auto|480p|720p|768p|1080p|2k|4k`. Default model is `seedance-fast`. SeeDance accepts native text-to-video with no image and integer output duration 4-15s (default 5s); `seedance-mini` supports 480p/720p and is best for cheaper drafts/multi-size tests; MiniMax H3 accepts native text-to-video, 4-15s output, public 768p/2k resolution, and up to 9 image, up to 3 video, and up to 3 audio references through Makaron Agent/chat. H3 defaults to 768p; request 2k explicitly for maximum/final quality. Kling supports 5-15s; Grok 1.5 supports 1-15s single-image-to-video only; Gemini Omni supports 3-10s fast 720p image/video generation and editing with native generated audio, including up to 6 image references when no video reference is provided. For `--video-model grok`, forced `--aspect` is ignored to avoid xAI stretching the source image; pad/create the image at the target shape first or use another model.
 
 Seedance 2.5: use `--video-model seedance-2.5` for 4-30 second output at 480p/720p. `--image` accepts local files or URLs (up to 30), while repeatable `--video` and `--audio` accept up to 10 each. Use `--video-operation generate|edit|extend`, `--extend-direction forward|backward`, `--output-format mp4|mov`, `--web-search`, `--generated-audio` / `--no-generated-audio`, and `--relaxed-content-filter`. Edit/extend require a video reference. The Evolink route does not currently expose 4K output.
 

package/bin/makaron.mjs

@@ -28,6 +28,14 @@ const NPM_PACKAGE_NAME = 'makaron-cli';
 const UPDATE_CHECK_INTERVAL_MS = 24 * 60 * 60 * 1000;
 const UPDATE_CHECK_TIMEOUT_MS = 400;
 const AGENT_WAIT_TIMEOUT_SECONDS = Math.max(900, Number(process.env.MAKARON_AGENT_WAIT_TIMEOUT_SECONDS || 10_800));
+const CHAT_AGENT_MODELS = [
+  'auto',
+  'gpt-5.6-terra',
+  'gpt-5.6-sol',
+  'gpt-5.6-luna',
+  'grok-4.5',
+  'deepseek-v4-pro',
+];
 
 // Public anon key (safe to embed — only enables auth, not data access)
 const SUPABASE_URL = 'https://sdyrtztrjgmmpnirswxt.supabase.co';
@@ -59,6 +67,16 @@ function warnLegacyModelFlag(replacement) {
   process.stderr.write(`⚠️  --model is deprecated here; use ${replacement}.\n`);
 }
 
+function validateChatAgentModel(value) {
+  if (!value || !CHAT_AGENT_MODELS.includes(value)) {
+    process.stderr.write(`❌ Unknown Agent LLM: ${value || '(missing value)'}\n`);
+    process.stderr.write(`Choose one of: ${CHAT_AGENT_MODELS.join(', ')}\n`);
+    process.stderr.write('--agent-model selects only the Agent LLM. Put image/video model preferences in the chat prompt, or use the explicit edit/video commands.\n');
+    process.exit(1);
+  }
+  return value;
+}
+
 function getCliVersion() {
   try {
     const pkg = JSON.parse(fs.readFileSync(new URL('../package.json', import.meta.url), 'utf-8'));
@@ -351,13 +369,15 @@ Options:
   --audio <file|url>        Attach a song, beat, or voice reference. MP3/WAV, repeatable.
   --media-manifest <file|-> Import source_url + start + end + description clips before this run.
   --skill <id|label|name>   Use an installed skill or auto-install a matched marketplace skill.
+  --agent-model <id>        Agent LLM only: auto, gpt-5.6-terra, gpt-5.6-sol,
+                            gpt-5.6-luna, grok-4.5, or deepseek-v4-pro.
   --background, -b          Submit and print a runId.
   --json                    Output structured JSON.
   --stream                  Legacy live SSE stream.
   --help, -h                Show this help.
 
-Model routing is automatic in chat. Do not pass --agent-model, --image-model,
---video-model, or the legacy --model flag.
+Agent LLM defaults to auto (currently gpt-5.6-terra). Image/video model routing
+stays automatic in chat; --image-model, --video-model, and --model are rejected.
 
 What you can ask:
   Image edit
@@ -381,6 +401,9 @@ What you can ask:
   Agent-to-agent source-range handoff
     makaron chat --project auto --media-manifest set-01.json -b --json "make a 30s vertical video"
 
+  Compare Agent LLMs with identical inputs
+    makaron chat --project auto --agent-model deepseek-v4-pro -b --json "make a 20s badminton video"
+
   Music
     makaron chat --project <id> "add calm piano background music"
 
@@ -422,6 +445,7 @@ async function streamAgent(baseUrl, headers, projectId, prompt, opts = {}) {
       projectId,
       prompt,
       headless: true,
+      ...(opts.agentModel ? { agentModel: opts.agentModel } : {}),
       ...(opts.uploadedVideoCount ? { uploadedVideoCount: opts.uploadedVideoCount } : {}),
       ...(opts.turnMediaCount ? { turnMediaCount: opts.turnMediaCount } : {}),
     }),
@@ -530,6 +554,7 @@ async function streamAgent(baseUrl, headers, projectId, prompt, opts = {}) {
 
 async function submitRun(baseUrl, headers, projectId, prompt, opts = {}) {
   const body = { projectId, prompt };
+  if (opts.agentModel) body.agentModel = opts.agentModel;
   if (opts.currentSnapshotIndex != null) body.currentSnapshotIndex = opts.currentSnapshotIndex;
   if (opts.isNsfw) body.isNsfw = opts.isNsfw;
   if (opts.audioAttachments?.length) body.audioAttachments = opts.audioAttachments;
@@ -1693,6 +1718,8 @@ Commands:
 
   chat --project <id> "message"      Chat (non-blocking, polls for result)
   chat --project <id> --skill <id>   Use a built-in or marketplace skill
+  chat --project <id> --agent-model <id> "message"
+                                     Select only the Agent LLM (strict allowlist)
   chat --project <id> --video <file> Attach video to conversation
   chat --project <id> --audio <file> Attach song/beat/voice reference
   chat --project auto --media-manifest <file> "message"
@@ -1726,7 +1753,8 @@ Examples:
   makaron chat --project <id> "turn this into a short video"
 
 Run makaron <command> --help for command-specific options.
-Chat chooses agent, image, and video models automatically.
+Chat defaults the Agent LLM automatically; --agent-model can select an exact
+Agent LLM. Image and video model routing remains automatic in chat.
 
 Environment:
   MAKARON_API_KEY       API key (mk_live_xxx) — recommended for agents
@@ -1854,7 +1882,7 @@ Use with chat:
     else console.log(`Video commands:
   video script --image <file> [--image <file>] "direction"   Write video script
   video create --script "..." --video-model seedance-fast    Native text-to-video (no image required)
-  video create --script "..." --video-model minimax-h3 --video-resolution 2k    MiniMax H3 native 2K text-to-video
+  video create --script "..." --video-model minimax-h3                          MiniMax H3 text-to-video (default 768P)
   video create --script "..." --image <url> [--duration 10]  Submit video task
   video create --script "..." --video <public-url> [--video-model seedance-fast|seedance-mini|seedance|seedance-2.5|kling|google-omni|minimax-h3]  Edit/reference a video (Grok does not support video refs)
   video status <taskId>                                      Check video status
@@ -1968,6 +1996,7 @@ if (!command || command === '--help' || command === '-h' || command === 'help')
   let jsonOutput = false;
   let activeSkill = undefined;
   let mediaManifestPath = undefined;
+  let agentModel = undefined;
   for (let i = 1; i < args.length; i++) {
     if (args[i] === '--project' && args[i + 1]) projectId = args[++i];
     else if (args[i] === '--image' && args[i + 1]) chatImages.push(args[++i]);
@@ -1983,6 +2012,21 @@ if (!command || command === '--help' || command === '-h' || command === 'help')
     else if (args[i].startsWith('--media-manifest=')) mediaManifestPath = args[i].slice('--media-manifest='.length);
     else if (args[i] === '--skill' && args[i + 1]) activeSkill = args[++i];
     else if (args[i].startsWith('--skill=')) activeSkill = args[i].slice('--skill='.length);
+    else if (args[i] === '--agent-model') {
+      if (agentModel !== undefined) {
+        process.stderr.write('❌ Pass --agent-model only once per chat run.\n');
+        process.exit(1);
+      }
+      if (!args[i + 1] || args[i + 1].startsWith('--')) validateChatAgentModel('');
+      agentModel = validateChatAgentModel(args[++i]);
+    }
+    else if (args[i].startsWith('--agent-model=')) {
+      if (agentModel !== undefined) {
+        process.stderr.write('❌ Pass --agent-model only once per chat run.\n');
+        process.exit(1);
+      }
+      agentModel = validateChatAgentModel(args[i].slice('--agent-model='.length));
+    }
     else if (args[i] === '--stream') useStream = true;
     else if (args[i] === '--background' || args[i] === '-b') background = true;
     else if (args[i] === '--json') jsonOutput = true;
@@ -1991,11 +2035,11 @@ if (!command || command === '--help' || command === '-h' || command === 'help')
       process.exit(1);
     }
     else if (
-      ['--agent-model', '--image-model', '--video-model', '--model'].includes(args[i])
-      || ['--agent-model=', '--image-model=', '--video-model=', '--model='].some(prefix => args[i].startsWith(prefix))
+      ['--image-model', '--video-model', '--model'].includes(args[i])
+      || ['--image-model=', '--video-model=', '--model='].some(prefix => args[i].startsWith(prefix))
     ) {
       const flag = args[i].split('=')[0];
-      process.stderr.write(`❌ makaron chat chooses agent, image, and video models automatically. Remove ${flag} and retry.\n`);
+      process.stderr.write(`❌ ${flag} is not valid for makaron chat. Only --agent-model may select the Agent LLM; image/video routing stays automatic. Remove ${flag} and retry.\n`);
       process.exit(1);
     }
     else promptParts.push(args[i]);
@@ -2222,6 +2266,7 @@ if (!command || command === '--help' || command === '-h' || command === 'help')
   if (useStream) {
     // Legacy SSE mode
     const { results } = await streamAgent(baseUrl, headers, projectId, finalPrompt, {
+      agentModel,
       uploadedVideoCount: uploadedTurnVideoCount,
       turnMediaCount: uploadedTurnMediaCount,
     });
@@ -2234,6 +2279,7 @@ if (!command || command === '--help' || command === '-h' || command === 'help')
   } else {
     // Default: fire-and-forget + poll
     const { runId } = await submitRun(baseUrl, headers, projectId, finalPrompt, {
+      agentModel,
       audioAttachments,
       uploadedVideoCount: uploadedTurnVideoCount,
       turnMediaCount: uploadedTurnMediaCount,
@@ -2776,7 +2822,7 @@ if (!command || command === '--help' || command === '-h' || command === 'help')
     console.log(`Video commands:
   video script --image <file> [--image <file>] "direction"   Write video script
   video create --script "..." --video-model seedance-fast    Native text-to-video (no image required)
-  video create --script "..." --video-model minimax-h3 --video-resolution 2k    MiniMax H3 native 2K text-to-video
+  video create --script "..." --video-model minimax-h3                          MiniMax H3 text-to-video (default 768P)
   video create --script "..." --image <url> [--duration 10]  Submit video task
   video create --script "..." --video <public-url> [--video-model seedance-fast|seedance-mini|seedance|seedance-2.5|kling|google-omni|minimax-h3]  Edit/reference a video (Grok does not support video refs)
   video status <taskId>                                      Check video status

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "makaron-cli",
-  "version": "0.13.6",
+  "version": "0.13.7",
   "description": "Talk to Makaron Agent from the terminal — create projects, edit images, generate videos",
   "type": "module",
   "scripts": {

package/skills/makaron/SKILL.md

@@ -85,7 +85,11 @@ npx makaron-cli chat --project auto --image photo.jpg --json -b "make it cinemat
 npx makaron-cli chat --project auto --image img1.jpg --image img2.jpg --json -b "combine these"
 ```
 
-`chat` always routes agent, image, and video models automatically. Never pass `--agent-model`, `--image-model`, `--video-model`, or the legacy `--model` flag to `chat`; the CLI rejects them before starting a run. Model flags remain available only on explicit low-level commands such as `edit` and `video create`.
+`chat` routes image and video models automatically. Use `--agent-model` only when the user explicitly asks to select or compare the reasoning/tool-calling Agent LLM. Accepted values are exactly `auto`, `gpt-5.6-terra`, `gpt-5.6-sol`, `gpt-5.6-luna`, `grok-4.5`, and `deepseek-v4-pro`; `auto` currently resolves to `gpt-5.6-terra`. Never put an image or video model ID in `--agent-model`. The CLI rejects unknown Agent IDs plus `--image-model`, `--video-model`, and legacy `--model` before starting a chat run.
+
+```bash
+npx makaron-cli chat --project auto --agent-model deepseek-v4-pro --json -b "make a 20s badminton video"
+```
 
 Returns immediately:
 ```json
@@ -280,7 +284,7 @@ npx makaron-cli video create --script "Shot 1 (15s): <<<image_1>>> and <<<image_
 
 # 3b. Native SeeDance or MiniMax H3 text-to-video (no image required)
 npx makaron-cli video create --script "Shot 1 (5s): A neon one-person studio wakes at dawn" --duration 5 --video-model seedance-fast --aspect 16:9
-npx makaron-cli video create --script "Shot 1 (15s): A premium creative editor comes alive" --duration 15 --video-model minimax-h3 --video-resolution 2k --aspect 16:9
+npx makaron-cli video create --script "Shot 1 (15s): A premium creative editor comes alive" --duration 15 --video-model minimax-h3 --aspect 16:9
 
 # 3c. Edit a video from a local file or public URL
 npx makaron-cli video create --script "make it funny" --video input.mp4 --duration 5 --video-model seedance-fast
@@ -296,7 +300,7 @@ npx makaron-cli video status <taskId>
 npx makaron-cli chat --project <id|auto> --video input.mp4 -b "make it funny"
 ```
 
-Options for `video create`: `--script "..."`, `--script-file <path>`, `--image <url>` (repeatable, up to the selected model limit), `--video <file|url>` and `--audio <file|url>` (repeatable where supported), `--duration <seconds>`, `--aspect 9:16|16:9|1:1`, `--video-model seedance-fast|seedance-mini|seedance|seedance-2.5|kling|grok|google-omni|minimax-h3`, `--video-resolution auto|480p|720p|768p|1080p|2k|4k`. Default model is `seedance-fast`. SeeDance accepts native text-to-video with no image and integer output duration 4-15s (default 5s); `seedance-mini` supports 480p/720p and is best for cheaper drafts/multi-size tests; MiniMax H3 accepts native text-to-video, 4-15s output, and up to 9 image, up to 3 video, and up to 3 audio references through Makaron Agent/chat. H3 defaults to public 2k output; 768p requires provider preview access plus `MINIMAX_H3_ENABLE_768P=true` on the Makaron server. Kling supports 5-15s; Grok 1.5 supports 1-15s single-image-to-video only; Gemini Omni supports 3-10s fast 720p image/video generation and editing with native generated audio, including up to 6 image references when no video reference is provided. For `--video-model grok`, forced `--aspect` is ignored to avoid xAI stretching the source image; pad/create the image at the target shape first or use another model.
+Options for `video create`: `--script "..."`, `--script-file <path>`, `--image <url>` (repeatable, up to the selected model limit), `--video <file|url>` and `--audio <file|url>` (repeatable where supported), `--duration <seconds>`, `--aspect 9:16|16:9|1:1`, `--video-model seedance-fast|seedance-mini|seedance|seedance-2.5|kling|grok|google-omni|minimax-h3`, `--video-resolution auto|480p|720p|768p|1080p|2k|4k`. Default model is `seedance-fast`. SeeDance accepts native text-to-video with no image and integer output duration 4-15s (default 5s); `seedance-mini` supports 480p/720p and is best for cheaper drafts/multi-size tests; MiniMax H3 accepts native text-to-video, 4-15s output, public 768p/2k resolution, and up to 9 image, up to 3 video, and up to 3 audio references through Makaron Agent/chat. H3 defaults to 768p; request 2k explicitly for maximum/final quality. Kling supports 5-15s; Grok 1.5 supports 1-15s single-image-to-video only; Gemini Omni supports 3-10s fast 720p image/video generation and editing with native generated audio, including up to 6 image references when no video reference is provided. For `--video-model grok`, forced `--aspect` is ignored to avoid xAI stretching the source image; pad/create the image at the target shape first or use another model.
 
 Seedance 2.5 uses `--video-model seedance-2.5` and supports 4-30s at 480p/720p, up to 30 images + 10 videos + 10 audios, repeatable local/URL references, `--video-operation generate|edit|extend`, `--extend-direction`, `--output-format mp4|mov`, and `--web-search`. The Evolink route does not currently expose 4K output.