opencode-model-router 1.0.0 → 1.0.2

package/README.md

@@ -14,9 +14,33 @@ The plugin injects a **delegation protocol** into the system prompt that teaches
 
 The agent automatically delegates via the Task tool when it recognizes the task complexity, or when plan steps are annotated with `[tier:fast]`, `[tier:medium]`, or `[tier:heavy]` tags.
 
+This applies both to plan-driven execution and direct ad-hoc requests. For every new user message, the orchestrator performs an intent gate, splits multi-task requests into atomic units, and routes each unit to `@fast`, `@medium`, or `@heavy`.
+
 ## Installation
 
-### Option A: Local plugin (recommended)
+### Option A: npm package (recommended)
+
+Add the plugin package in your `opencode.json`:
+
+```json
+{
+  "plugin": [
+    "opencode-model-router@1.0.0"
+  ]
+}
+```
+
+If you prefer always getting the latest release, use:
+
+```json
+{
+  "plugin": [
+    "opencode-model-router"
+  ]
+}
+```
+
+### Option B: Local plugin clone
 
 Clone directly into your OpenCode plugins directory:
 
@@ -35,7 +59,7 @@ Then add it to your `opencode.json`:
 }
 ```
 
-### Option B: Reference from anywhere
+### Option C: Reference from anywhere
 
 Clone wherever you want:
 
@@ -74,8 +98,8 @@ The plugin ships with two presets:
 | Tier | Model | Notes |
 |------|-------|-------|
 | fast | `openai/gpt-5.3-codex-spark` | Cheapest, fastest |
-| medium | `openai/o3` | Medium reasoning effort |
-| heavy | `openai/o3` | High reasoning effort |
+| medium | `openai/gpt-5.3-codex` | Default settings (no variant/reasoning override) |
+| heavy | `openai/gpt-5.3-codex` | Variant: `xhigh` |
 
 Switch presets with the `/preset` command:
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-model-router",
-  "version": "1.0.0",
+  "version": "1.0.2",
   "description": "OpenCode plugin that routes tasks to tiered subagents (fast/medium/heavy) based on complexity",
   "type": "module",
   "main": "./src/index.ts",

package/src/index.ts

@@ -112,6 +112,21 @@ function buildDelegationProtocol(cfg: RouterConfig): string {
 
   const rules = cfg.rules.map((r, i) => `${i + 1}. ${r}`).join("\n");
 
+  const directRoutingRules = [
+    "Run an Intent Gate for every new user message, even without a plan file.",
+    "If the message contains multiple asks, split it into atomic tasks before routing.",
+    "Route read-only exploration/search tasks to @fast.",
+    "Route implementation/edit/test/refactor tasks to @medium.",
+    "Route architecture/security/performance/complex debugging tasks to @heavy.",
+    "If a request mixes tiers, delegate each subtask to the right tier and then synthesize one final response.",
+    "If a task is trivial (single grep/read), execute directly instead of delegating.",
+    "When uncertain, default to the configured default tier unless user constraints clearly require another tier.",
+  ];
+
+  const directRoutingList = directRoutingRules
+    .map((rule, index) => `${index + 1}. ${rule}`)
+    .join("\n");
+
   return [
     "## Model Delegation Protocol",
     "",
@@ -128,6 +143,10 @@ function buildDelegationProtocol(cfg: RouterConfig): string {
     "### Rules:",
     rules,
     "",
+    "### Direct Request Routing (No Plan Required):",
+    "These rules also apply to ad-hoc user requests, not only PLAN.md execution.",
+    directRoutingList,
+    "",
     "### How to delegate:",
     "Use the Task tool with the tier name as `subagent_type`:",
     '- `Task(subagent_type="fast", prompt="Find all files importing AuthContext")`',

package/tiers.json

@@ -55,9 +55,8 @@
         ]
       },
       "medium": {
-        "model": "openai/o3",
-        "reasoning": { "effort": "medium", "summary": "auto" },
-        "description": "o3 medium reasoning for implementation and standard coding",
+        "model": "openai/gpt-5.3-codex",
+        "description": "GPT-5.3 Codex default settings for implementation and standard coding",
         "steps": 50,
         "prompt": "You are an implementation agent. Write clean, production-quality code matching existing project patterns. Run linters/tests after changes when possible.",
         "whenToUse": [
@@ -68,9 +67,9 @@
         ]
       },
       "heavy": {
-        "model": "openai/o3",
-        "reasoning": { "effort": "high", "summary": "detailed" },
-        "description": "o3 high reasoning for architecture and complex tasks",
+        "model": "openai/gpt-5.3-codex",
+        "variant": "xhigh",
+        "description": "GPT-5.3 Codex xhigh for architecture and complex tasks",
         "steps": 30,
         "prompt": "You are a senior architecture consultant. Analyze deeply, consider tradeoffs, and provide thorough reasoning.",
         "whenToUse": [