opencode-model-router 1.1.4 → 1.1.6

package/README.md

@@ -23,8 +23,8 @@ A keyword routing guide (`@fast→search/grep/read`, `@medium→impl/refactor/te
 **Skip delegation overhead for trivial work.**
 Single grep? One file read? The orchestrator executes directly — zero delegation cost, zero latency.
 
-**Three routing modes for different budgets.**
-`/budget normal` (balanced), `/budget budget` (aggressive savings, defaults everything to @fast), `/budget quality` (liberal use of stronger models). Mode persists across restarts.
+**Four routing modes for different budgets.**
+`/budget normal` (balanced), `/budget budget` (aggressive savings, defaults everything to @fast), `/budget quality` (liberal use of stronger models), `/budget deep` (heavy-first for long architecture/debug runs). Mode persists across restarts.
 
 **Cost ratios in the prompt.**
 Every tier carries its `costRatio` (fast=1x, medium=5x, heavy=20x) injected into the system prompt. The orchestrator sees the price before deciding. It picks the cheapest tier that can reliably handle the task.
@@ -278,6 +278,7 @@ Switch with `/budget <mode>`. Mode is persisted across restarts.
 | `normal` | @medium | Balanced — routes by task complexity |
 | `budget` | @fast | Aggressive savings — defaults cheap, escalates only when necessary |
 | `quality` | @medium | Quality-first — liberal use of @medium/@heavy |
+| `deep` | @heavy | Deep-analysis mode — heavy-first for architecture/debug/security with longer heavy runs |
 
 ```json
 {
@@ -286,15 +287,28 @@ Switch with `/budget <mode>`. Mode is persisted across restarts.
       "defaultTier": "fast",
       "description": "Aggressive cost savings",
       "overrideRules": [
-        "Default ALL tasks to @fast unless they clearly require code edits",
-        "Use @medium ONLY for: multi-file edits, complex refactors, test suites",
-        "Use @heavy ONLY when explicitly requested or after 2+ failed @medium attempts"
+        "default→@fast unless edits/complex-reasoning needed",
+        "@medium ONLY: multi-file-edit/refactor/test-suite/build-fix",
+        "@heavy ONLY: user-requested OR ≥2 @medium failures"
+      ]
+    },
+    "deep": {
+      "defaultTier": "heavy",
+      "description": "Deep analysis mode — prioritizes thorough architecture/debug work with long heavy runs",
+      "overrideRules": [
+        "default→@medium for implementation and multi-file changes",
+        "@heavy for architecture/debug/security/tradeoff-analysis by default",
+        "allow long heavy runs before fallback; avoid premature downshift",
+        "trivial(grep/read/glob)→direct,no-delegate",
+        "if task is composite: explore@fast then execute@heavy"
       ]
     }
   }
 }
 ```
 
+**Heavy tool-call budget:** `@heavy.steps=120` by default across presets (raised from 60) to reduce premature cutoffs on long architecture/debug tasks.
+
 ### Task taxonomy (`taskPatterns`)
 
 Keyword routing guide injected into the system prompt. Customize to match your workflow:
@@ -381,7 +395,7 @@ Defines provider fallback order when a delegated task fails:
 | `/preset` | List available presets |
 | `/preset <name>` | Switch preset (e.g., `/preset openai`) |
 | `/budget` | Show available modes and which is active |
-| `/budget <mode>` | Switch routing mode (`normal`, `budget`, `quality`) |
+| `/budget <mode>` | Switch routing mode (`normal`, `budget`, `quality`, `deep`) |
 | `/annotate-plan [path]` | Annotate a plan file with `[tier:X]` tags for each step |
 
 ## Plan annotation

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-model-router",
-  "version": "1.1.4",
+  "version": "1.1.6",
   "description": "OpenCode plugin that routes tasks to tiered subagents (fast/medium/heavy) based on complexity",
   "type": "module",
   "main": "./src/index.ts",

package/tiers.json

@@ -36,7 +36,7 @@
         "variant": "max",
         "costRatio": 20,
         "description": "Opus 4.6 max for architecture, complex debugging, and security",
-        "steps": 30,
+        "steps": 120,
         "prompt": "You are a senior architecture consultant. Analyze deeply, consider tradeoffs, and provide thorough reasoning. Be exhaustive in your analysis.",
         "whenToUse": [
           "Architecture decisions",
@@ -77,7 +77,7 @@
         "variant": "xhigh",
         "costRatio": 20,
         "description": "GPT-5.3 Codex xhigh for architecture and complex tasks",
-        "steps": 30,
+        "steps": 120,
         "prompt": "You are a senior architecture consultant. Analyze deeply, consider tradeoffs, and provide thorough reasoning.",
         "whenToUse": [
           "Architecture decisions",
@@ -120,7 +120,7 @@
         "variant": "thinking",
         "costRatio": 20,
         "description": "Claude Opus 4.6 via GitHub Copilot for architecture, complex debugging, and security",
-        "steps": 30,
+        "steps": 120,
         "prompt": "You are a senior architecture consultant. Analyze deeply, consider tradeoffs, and provide thorough reasoning. Be exhaustive in your analysis.",
         "whenToUse": [
           "Architecture decisions",
@@ -162,7 +162,51 @@
         "model": "google/gemini-3-pro-preview",
         "costRatio": 20,
         "description": "Gemini 3 Pro Preview for architecture, complex debugging, and security",
+        "steps": 120,
+        "prompt": "You are a senior architecture consultant. Analyze deeply, consider tradeoffs, and provide thorough reasoning. Be exhaustive in your analysis.",
+        "whenToUse": [
+          "Architecture decisions",
+          "Complex debugging (after 2+ failures)",
+          "Security review",
+          "Performance optimization"
+        ]
+      }
+    },
+    "hybrid": {
+      "fast": {
+        "model": "openai/gpt-5.3-codex-spark",
+        "costRatio": 1,
+        "description": "GPT-5.3 Codex Spark for exploration, search, and simple reads",
         "steps": 30,
+        "prompt": "You are a fast exploration agent. Focus on speed and efficiency. Read files, search code, and return findings concisely. Do NOT make edits unless explicitly asked.",
+        "whenToUse": [
+          "Codebase exploration and search",
+          "Simple file reads and listing",
+          "Grep/glob operations",
+          "Quick lookups and research"
+        ]
+      },
+      "medium": {
+        "model": "anthropic/claude-sonnet-4-6",
+        "variant": "max",
+        "costRatio": 5,
+        "description": "Claude Sonnet 4.6 max for implementation, refactoring, and tests",
+        "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": [
+          "Feature implementation",
+          "Refactoring",
+          "Writing tests",
+          "Code review",
+          "Bug fixes"
+        ]
+      },
+      "heavy": {
+        "model": "anthropic/claude-opus-4-6",
+        "variant": "max",
+        "costRatio": 20,
+        "description": "Claude Opus 4.6 max for architecture, complex debugging, and security",
+        "steps": 120,
         "prompt": "You are a senior architecture consultant. Analyze deeply, consider tradeoffs, and provide thorough reasoning. Be exhaustive in your analysis.",
         "whenToUse": [
           "Architecture decisions",
@@ -235,6 +279,17 @@
         "@fast ONLY: trivial single-tool ops (1 grep/1 read)",
         "prefer thoroughness over speed"
       ]
+    },
+    "deep": {
+      "defaultTier": "heavy",
+      "description": "Deep analysis mode — prioritizes thorough architecture/debug work with long heavy runs",
+      "overrideRules": [
+        "default→@medium for implementation and multi-file changes",
+        "@heavy for architecture/debug/security/tradeoff-analysis by default",
+        "allow long heavy runs before fallback; avoid premature downshift",
+        "trivial(grep/read/glob)→direct,no-delegate",
+        "if task is composite: explore@fast then execute@heavy"
+      ]
     }
   },
   "fallback": {