cc-mirror 1.4.0 → 1.4.1

package/dist/skills/orchestration/SKILL.md

@@ -270,6 +270,79 @@ Create src/routes/auth.ts with:
 )
 ```
 
+### Model Selection
+
+Choose the right model for each agent's task:
+
+```
+┌─────────────────────────────────────────────────────────────┐
+│  HAIKU (model="haiku") — The Errand Runner                  │
+│                                                             │
+│  Spawn many of these. They're fast and cheap.               │
+│                                                             │
+│  • Fetch files, grep for patterns, find things              │
+│  • Simple lookups and searches                              │
+│  • Gather raw information for you to synthesize             │
+│  • Mechanical tasks with no judgment calls                  │
+│  • Run 5-10 in parallel to explore quickly                  │
+│                                                             │
+├─────────────────────────────────────────────────────────────┤
+│  SONNET (model="sonnet") — The Capable Worker               │
+│                                                             │
+│  Smart, but needs clear direction. Like a junior-mid dev.   │
+│                                                             │
+│  • Well-structured implementation tasks                     │
+│  • Research: reading docs, understanding APIs               │
+│  • Following established patterns in a codebase             │
+│  • Semi-difficult analysis with clear scope                 │
+│  • Test generation, documentation                           │
+│  • When the task is clear and you've defined what to do     │
+│                                                             │
+├─────────────────────────────────────────────────────────────┤
+│  OPUS (model="opus") — The Critical Thinker                 │
+│                                                             │
+│  Thinks for itself. Trust its judgment.                     │
+│                                                             │
+│  • Ambiguous or underspecified problems                     │
+│  • Architectural decisions and design trade-offs            │
+│  • Complex debugging requiring reasoning across systems     │
+│  • Security review, vulnerability assessment                │
+│  • When you need creative problem-solving                   │
+│  • Tasks where quality of thinking matters most             │
+│  • When the path forward isn't obvious                      │
+│                                                             │
+└─────────────────────────────────────────────────────────────┘
+```
+
+**Example with model selection:**
+
+```
+# Gather info - spawn haiku wildly
+Task(subagent_type="Explore", description="Find auth files", prompt="...", model="haiku", run_in_background=True)
+Task(subagent_type="Explore", description="Find user routes", prompt="...", model="haiku", run_in_background=True)
+Task(subagent_type="Explore", description="Find middleware", prompt="...", model="haiku", run_in_background=True)
+
+# Clear implementation task - sonnet
+Task(
+    subagent_type="general-purpose",
+    description="Implement login route",
+    prompt="Create POST /login following the pattern in src/routes/users.ts...",
+    model="sonnet",
+    run_in_background=True
+)
+
+# Needs judgment and critical thinking - opus
+Task(
+    subagent_type="general-purpose",
+    description="Design auth architecture",
+    prompt="Analyze the codebase and recommend the best auth approach...",
+    model="opus",
+    run_in_background=True
+)
+```
+
+**Always pass `model` explicitly.** Haiku for gathering, sonnet for well-defined work, opus when you need real thinking.
+
 ---
 
 ## 🚀 The Orchestration Flow

package/dist/skills/orchestration/references/domains/code-review.md

@@ -43,12 +43,17 @@ Phase 2: REDUCE (Synthesize)
 **Implementation:**
 
 ```
-# All in single message for parallelism
-Task(subagent_type="Explore", prompt="Fetch PR #123 details, understand context and related issues")
-Task(subagent_type="general-purpose", prompt="Review code quality: patterns, readability, maintainability")
-Task(subagent_type="general-purpose", prompt="Review logic: correctness, edge cases, error handling")
-Task(subagent_type="general-purpose", prompt="Review security: injection, auth, data exposure")
-Task(subagent_type="general-purpose", prompt="Review performance: complexity, queries, memory")
+# All in single message for parallelism (opus for reviews - critical thinking)
+Task(subagent_type="Explore", prompt="Fetch PR #123 details, understand context and related issues",
+     model="haiku", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="Review code quality: patterns, readability, maintainability",
+     model="opus", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="Review logic: correctness, edge cases, error handling",
+     model="opus", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="Review security: injection, auth, data exposure",
+     model="opus", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="Review performance: complexity, queries, memory",
+     model="opus", run_in_background=True)
 ```
 
 ### Pattern: Contextual Deep-Dive
@@ -287,11 +292,15 @@ TaskCreate(subject="Synthesize review", description="Aggregate findings into rev
 # Synthesis blocked by analysis tasks
 TaskUpdate(taskId="5", addBlockedBy=["1", "2", "3", "4"])
 
-# Spawn parallel agents for analysis
-Task(subagent_type="general-purpose", prompt="TaskId 1: Analyze PR context...")
-Task(subagent_type="general-purpose", prompt="TaskId 2: Review code quality...")
-Task(subagent_type="general-purpose", prompt="TaskId 3: Check security...")
-Task(subagent_type="general-purpose", prompt="TaskId 4: Assess performance...")
+# Spawn parallel agents for analysis (opus for reviews - critical thinking)
+Task(subagent_type="Explore", prompt="TaskId 1: Analyze PR context...",
+     model="haiku", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="TaskId 2: Review code quality...",
+     model="opus", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="TaskId 3: Check security...",
+     model="opus", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="TaskId 4: Assess performance...",
+     model="opus", run_in_background=True)
 ```
 
 ---

package/dist/skills/orchestration/references/domains/data-analysis.md

@@ -326,10 +326,13 @@ TaskUpdate(taskId="4", addBlockedBy=["1"])
 TaskUpdate(taskId="5", addBlockedBy=["2", "3", "4"])
 TaskUpdate(taskId="6", addBlockedBy=["5"])
 
-# Spawn parallel analysis agents
-Task(subagent_type="general-purpose", prompt="TaskId 2: Explore distributions...")
-Task(subagent_type="general-purpose", prompt="TaskId 3: Analyze missing data...")
-Task(subagent_type="general-purpose", prompt="TaskId 4: Check data quality...")
+# Spawn parallel analysis agents (haiku for data exploration)
+Task(subagent_type="Explore", prompt="TaskId 2: Explore distributions...",
+     model="haiku", run_in_background=True)
+Task(subagent_type="Explore", prompt="TaskId 3: Analyze missing data...",
+     model="haiku", run_in_background=True)
+Task(subagent_type="Explore", prompt="TaskId 4: Check data quality...",
+     model="haiku", run_in_background=True)
 ```
 
 ## Best Practices

package/dist/skills/orchestration/references/domains/devops.md

@@ -320,9 +320,11 @@ TaskUpdate(taskId="4", addBlockedBy=["2"])  # Can parallel with network
 TaskUpdate(taskId="5", addBlockedBy=["3", "4"])
 TaskUpdate(taskId="6", addBlockedBy=["5"])
 
-# Parallel infrastructure implementation
-Task(subagent_type="general-purpose", prompt="TaskId 3: Implement network changes...")
-Task(subagent_type="general-purpose", prompt="TaskId 4: Implement compute changes...")
+# Parallel infrastructure implementation (sonnet for well-structured work)
+Task(subagent_type="general-purpose", prompt="TaskId 3: Implement network changes...",
+     model="sonnet", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="TaskId 4: Implement compute changes...",
+     model="sonnet", run_in_background=True)
 ```
 
 ## Safety Principles

package/dist/skills/orchestration/references/domains/documentation.md

@@ -320,10 +320,13 @@ TaskUpdate(taskId="4", addBlockedBy=["1"])
 TaskUpdate(taskId="5", addBlockedBy=["2", "3", "4"])
 TaskUpdate(taskId="6", addBlockedBy=["5"])
 
-# Spawn parallel documentation agents
-Task(subagent_type="general-purpose", prompt="TaskId 2: Document API endpoints...")
-Task(subagent_type="general-purpose", prompt="TaskId 3: Document components...")
-Task(subagent_type="general-purpose", prompt="TaskId 4: Document utilities...")
+# Spawn parallel documentation agents (sonnet for well-structured work)
+Task(subagent_type="general-purpose", prompt="TaskId 2: Document API endpoints...",
+     model="sonnet", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="TaskId 3: Document components...",
+     model="sonnet", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="TaskId 4: Document utilities...",
+     model="sonnet", run_in_background=True)
 ```
 
 ## Output Formats

package/dist/skills/orchestration/references/domains/project-management.md

@@ -350,9 +350,11 @@ TaskUpdate(taskId="4", addBlockedBy=["2"])
 TaskUpdate(taskId="5", addBlockedBy=["3", "4"])
 TaskUpdate(taskId="6", addBlockedBy=["5"])
 
-# Parallel breakdown
-Task(subagent_type="general-purpose", prompt="TaskId 3: Break down Story A...")
-Task(subagent_type="general-purpose", prompt="TaskId 4: Break down Story B...")
+# Parallel breakdown (sonnet for structured planning work)
+Task(subagent_type="general-purpose", prompt="TaskId 3: Break down Story A...",
+     model="sonnet", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="TaskId 4: Break down Story B...",
+     model="sonnet", run_in_background=True)
 ```
 
 ## Best Practices

package/dist/skills/orchestration/references/domains/research.md

@@ -300,10 +300,13 @@ TaskUpdate(taskId="3", addBlockedBy=["1"])
 TaskUpdate(taskId="4", addBlockedBy=["1"])
 TaskUpdate(taskId="5", addBlockedBy=["2", "3", "4"])
 
-# Spawn Explore agents in parallel
-Task(subagent_type="Explore", prompt="TaskId 2: Find auth patterns...")
-Task(subagent_type="Explore", prompt="TaskId 3: Find API patterns...")
-Task(subagent_type="Explore", prompt="TaskId 4: Find database patterns...")
+# Spawn Explore agents in parallel (haiku swarm for fast exploration)
+Task(subagent_type="Explore", prompt="TaskId 2: Find auth patterns...",
+     model="haiku", run_in_background=True)
+Task(subagent_type="Explore", prompt="TaskId 3: Find API patterns...",
+     model="haiku", run_in_background=True)
+Task(subagent_type="Explore", prompt="TaskId 4: Find database patterns...",
+     model="haiku", run_in_background=True)
 ```
 
 ## Agent Selection for Research

package/dist/skills/orchestration/references/domains/software-development.md

@@ -256,8 +256,9 @@ TaskUpdate(taskId="3", addBlockedBy=["2"])  # Implement after design
 TaskUpdate(taskId="4", addBlockedBy=["3"])  # Error handling after core
 TaskUpdate(taskId="5", addBlockedBy=["3"])  # Tests can parallel with error handling
 
-# Spawn agents for unblocked tasks
-Task(subagent_type="general-purpose", prompt="TaskId 1: Analyze requirements...")
+# Spawn agents for unblocked tasks (haiku for analysis/exploration)
+Task(subagent_type="Explore", prompt="TaskId 1: Analyze requirements...",
+     model="haiku", run_in_background=True)
 ```
 
 Agents mark tasks resolved immediately upon completion.

package/dist/skills/orchestration/references/domains/testing.md

@@ -293,9 +293,11 @@ TaskUpdate(taskId="4", addBlockedBy=["2", "3"])
 TaskUpdate(taskId="5", addBlockedBy=["4"])
 TaskUpdate(taskId="6", addBlockedBy=["5"])
 
-# Parallel test generation
-Task(subagent_type="general-purpose", prompt="TaskId 2: Generate unit tests...")
-Task(subagent_type="general-purpose", prompt="TaskId 3: Generate integration tests...")
+# Parallel test generation (sonnet for well-structured work)
+Task(subagent_type="general-purpose", prompt="TaskId 2: Generate unit tests...",
+     model="sonnet", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="TaskId 3: Generate integration tests...",
+     model="sonnet", run_in_background=True)
 ```
 
 ## Test Execution Best Practices

package/dist/skills/orchestration/references/examples.md

@@ -36,16 +36,20 @@
 
 ### Execution
 
-```python
+```
 # Phase 1: Gather context + parallel analysis (single message)
 Task(subagent_type="Explore", description="Get PR context",
-     prompt="Fetch PR #123 details, understand what changed and why")
+     prompt="Fetch PR #123 details, understand what changed and why",
+     model="haiku", run_in_background=True)
 Task(subagent_type="general-purpose", description="Review quality",
-     prompt="Review code quality: readability, patterns, maintainability")
+     prompt="Review code quality: readability, patterns, maintainability",
+     model="opus", run_in_background=True)  # Critical thinking for review
 Task(subagent_type="general-purpose", description="Review security",
-     prompt="Review security: injection, auth, data exposure risks")
+     prompt="Review security: injection, auth, data exposure risks",
+     model="opus", run_in_background=True)  # Security needs judgment
 Task(subagent_type="general-purpose", description="Review performance",
-     prompt="Review performance: complexity, queries, memory usage")
+     prompt="Review performance: complexity, queries, memory usage",
+     model="opus", run_in_background=True)  # Performance analysis
 ```
 
 ### User Sees
@@ -91,26 +95,32 @@ Adds user profile editing with image upload capability.
 
 ### Execution
 
-```python
-# Phase 1: Research (Pipeline)
+```
+# Phase 1: Research (haiku for exploration)
 context = Task(subagent_type="Explore", description="Find styling patterns",
-               prompt="Find existing theme/styling patterns, CSS architecture")
+               prompt="Find existing theme/styling patterns, CSS architecture",
+               model="haiku", run_in_background=True)
 
-# Phase 2: Plan
+# Phase 2: Plan (opus for design decisions)
 plan = Task(subagent_type="Plan", description="Design dark mode",
-            prompt=f"Given: {context}. Design dark mode implementation.")
+            prompt=f"Given: {context}. Design dark mode implementation.",
+            model="opus", run_in_background=True)
 
-# Phase 3: Implement (Fan-Out - single message)
+# Phase 3: Implement (sonnet for well-structured work - single message)
 Task(subagent_type="general-purpose", description="Add CSS variables",
-     prompt="Add dark theme CSS variables...")
+     prompt="Add dark theme CSS variables...",
+     model="sonnet", run_in_background=True)
 Task(subagent_type="general-purpose", description="Create toggle",
-     prompt="Create theme toggle component...")
+     prompt="Create theme toggle component...",
+     model="sonnet", run_in_background=True)
 Task(subagent_type="general-purpose", description="Add persistence",
-     prompt="Add localStorage persistence for theme preference...")
+     prompt="Add localStorage persistence for theme preference...",
+     model="sonnet", run_in_background=True)
 
-# Phase 4: Integration (Pipeline)
+# Phase 4: Integration (sonnet for implementation)
 Task(subagent_type="general-purpose", description="Wire and test",
-     prompt="Wire components together, test theme switching works")
+     prompt="Wire components together, test theme switching works",
+     model="sonnet", run_in_background=True)
 ```
 
 ### User Sees
@@ -162,25 +172,30 @@ Try it: Click the moon/sun icon in the header.
 
 ### Execution
 
-```python
-# Phase 1: Parallel diagnosis (Fan-Out - single message)
+```
+# Phase 1: Parallel diagnosis (haiku swarm - single message)
 Task(subagent_type="Explore", description="Check logs",
-     prompt="Search for login errors, password reset issues in logs")
+     prompt="Search for login errors, password reset issues in logs",
+     model="haiku", run_in_background=True)
 Task(subagent_type="Explore", description="Find reset flow",
-     prompt="Find password reset implementation, trace the flow")
+     prompt="Find password reset implementation, trace the flow",
+     model="haiku", run_in_background=True)
 Task(subagent_type="Explore", description="Check session handling",
-     prompt="How are sessions/tokens handled after password change?")
+     prompt="How are sessions/tokens handled after password change?",
+     model="haiku", run_in_background=True)
 
 # Phase 2: Synthesize (after results return)
 # Identify: Password reset invalidates session but doesn't clear cookie
 
-# Phase 3: Fix
+# Phase 3: Fix (sonnet for implementation)
 Task(subagent_type="general-purpose", description="Fix session bug",
-     prompt="Clear session cookie after password reset...")
+     prompt="Clear session cookie after password reset...",
+     model="sonnet", run_in_background=True)
 
-# Phase 4: Verify
+# Phase 4: Verify (sonnet for test writing)
 Task(subagent_type="general-purpose", description="Add regression test",
-     prompt="Add test for login after password reset...")
+     prompt="Add test for login after password reset...",
+     model="sonnet", run_in_background=True)
 ```
 
 ### User Sees
@@ -229,14 +244,17 @@ The flow now:
 
 ### Execution
 
-```python
-# Phase 1: Parallel exploration (Fan-Out - single message)
+```
+# Phase 1: Parallel exploration (haiku swarm - single message)
 Task(subagent_type="Explore", description="Project structure",
-     prompt="Analyze project structure, entry points, build system")
+     prompt="Analyze project structure, entry points, build system",
+     model="haiku", run_in_background=True)
 Task(subagent_type="Explore", description="Core patterns",
-     prompt="Identify core architectural patterns, frameworks used")
+     prompt="Identify core architectural patterns, frameworks used",
+     model="haiku", run_in_background=True)
 Task(subagent_type="Explore", description="Data flow",
-     prompt="Trace main data flows, API structure, state management")
+     prompt="Trace main data flows, API structure, state management",
+     model="haiku", run_in_background=True)
 ```
 
 ### User Sees
@@ -294,27 +312,31 @@ Want me to dive deeper into any area?
 
 ### Execution
 
-```python
-# Phase 1: Run tests in background
+```
+# Phase 1: Run tests in background (sonnet for test execution)
 Task(subagent_type="general-purpose", description="Run tests",
      prompt="Run full test suite, report failures",
-     run_in_background=True)
+     model="sonnet", run_in_background=True)
 
 # Continue with other work or wait
 result = TaskOutput(task_id="...", block=True)
 
-# Phase 2: Fix failures (Fan-Out if multiple)
+# Phase 2: Fix failures (sonnet for fixes - single message)
 # If 3 failures found:
 Task(subagent_type="general-purpose", description="Fix test 1",
-     prompt="Fix failing test in auth.test.ts...")
+     prompt="Fix failing test in auth.test.ts...",
+     model="sonnet", run_in_background=True)
 Task(subagent_type="general-purpose", description="Fix test 2",
-     prompt="Fix failing test in api.test.ts...")
+     prompt="Fix failing test in api.test.ts...",
+     model="sonnet", run_in_background=True)
 Task(subagent_type="general-purpose", description="Fix test 3",
-     prompt="Fix failing test in utils.test.ts...")
+     prompt="Fix failing test in utils.test.ts...",
+     model="sonnet", run_in_background=True)
 
-# Phase 3: Verify
+# Phase 3: Verify (sonnet for test execution)
 Task(subagent_type="general-purpose", description="Re-run tests",
-     prompt="Run test suite again to verify fixes")
+     prompt="Run test suite again to verify fixes",
+     model="sonnet", run_in_background=True)
 ````
 
 ### User Sees

package/dist/skills/orchestration/references/patterns.md

@@ -66,7 +66,7 @@ TaskUpdate(taskId="3", addBlockedBy=["2"])
 
 # 3. Find unblocked tasks and spawn agents
 TaskList()  # Find tasks with empty blockedBy
-Task(subagent_type="general-purpose", prompt="TaskId 1: Setup database...", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="TaskId 1: Setup database...", model="sonnet", run_in_background=True)
 
 # 4. Agents mark complete, repeat
 TaskUpdate(taskId="1", status="resolved")
@@ -91,10 +91,10 @@ Orchestrator
 **Implementation:**
 
 ```python
-# Single message with multiple background agents
-Task(subagent_type="Explore", prompt="Analyze auth module...", run_in_background=True)
-Task(subagent_type="Explore", prompt="Analyze database layer...", run_in_background=True)
-Task(subagent_type="Explore", prompt="Analyze API routes...", run_in_background=True)
+# Single message with multiple background agents (haiku for fast exploration)
+Task(subagent_type="Explore", prompt="Analyze auth module...", model="haiku", run_in_background=True)
+Task(subagent_type="Explore", prompt="Analyze database layer...", model="haiku", run_in_background=True)
+Task(subagent_type="Explore", prompt="Analyze API routes...", model="haiku", run_in_background=True)
 ```
 
 **Critical:** All Task calls MUST be in ONE message AND use `run_in_background=True`.
@@ -114,16 +114,16 @@ Agent A → output → Agent B → output → Agent C → final result
 **Implementation:**
 
 ```python
-# Step 1: Research (background, wait for notification)
-Task(subagent_type="Explore", prompt="Find all API endpoints...", run_in_background=True)
+# Step 1: Research (haiku - fast exploration)
+Task(subagent_type="Explore", prompt="Find all API endpoints...", model="haiku", run_in_background=True)
 # → Notification arrives with result1
 
-# Step 2: Plan (background, uses result1)
-Task(subagent_type="Plan", prompt=f"Given endpoints: {result1}, design...", run_in_background=True)
+# Step 2: Plan (opus - needs critical thinking for design)
+Task(subagent_type="Plan", prompt=f"Given endpoints: {result1}, design...", model="opus", run_in_background=True)
 # → Notification arrives with result2
 
-# Step 3: Implement (background, uses result2)
-Task(subagent_type="general-purpose", prompt=f"Implement this plan: {result2}", run_in_background=True)
+# Step 3: Implement (sonnet - well-structured from plan)
+Task(subagent_type="general-purpose", prompt=f"Implement this plan: {result2}", model="sonnet", run_in_background=True)
 ```
 
 ---
@@ -143,10 +143,10 @@ Input ──►├──► Agent B ──┼──► Aggregator → Final Resu
 **Implementation:**
 
 ```python
-# MAP: Launch parallel agents (single message)
-Task(subagent_type="general-purpose", prompt="Analyze file1.ts for security issues", run_in_background=True)
-Task(subagent_type="general-purpose", prompt="Analyze file2.ts for security issues", run_in_background=True)
-Task(subagent_type="general-purpose", prompt="Analyze file3.ts for security issues", run_in_background=True)
+# MAP: Launch parallel agents (opus for security - needs critical thinking)
+Task(subagent_type="general-purpose", prompt="Analyze file1.ts for security issues", model="opus", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="Analyze file2.ts for security issues", model="opus", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="Analyze file3.ts for security issues", model="opus", run_in_background=True)
 
 # REDUCE: Collect and synthesize
 results = [TaskOutput(task_id=id) for id in task_ids]
@@ -170,10 +170,10 @@ Problem ─├──► Approach B ──┼──► Evaluate → Best Solution
 **Implementation:**
 
 ```python
-# Launch competing approaches (single message, all background)
-Task(subagent_type="general-purpose", prompt="Implement using recursive approach...", run_in_background=True)
-Task(subagent_type="general-purpose", prompt="Implement using iterative approach...", run_in_background=True)
-Task(subagent_type="general-purpose", prompt="Implement using memoization...", run_in_background=True)
+# Launch competing approaches (sonnet for implementation)
+Task(subagent_type="general-purpose", prompt="Implement using recursive approach...", model="sonnet", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="Implement using iterative approach...", model="sonnet", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="Implement using memoization...", model="sonnet", run_in_background=True)
 
 # Notifications arrive → Evaluate and select best
 ```
@@ -189,8 +189,8 @@ Long-running agents while continuing foreground work.
 **Implementation:**
 
 ```python
-# Launch background work
-Task(subagent_type="general-purpose", prompt="Run full test suite...", run_in_background=True)
+# Launch background work (sonnet for test execution)
+Task(subagent_type="general-purpose", prompt="Run full test suite...", model="sonnet", run_in_background=True)
 
 # Continue foreground work
 # ... do other tasks ...
@@ -271,13 +271,13 @@ results = [TaskOutput(id) for id in task_ids]
 ### Speculative + Pipeline
 
 ```python
-# Try multiple approaches (background)
-Task(subagent_type="general-purpose", prompt="Approach A...", run_in_background=True)
-Task(subagent_type="general-purpose", prompt="Approach B...", run_in_background=True)
+# Try multiple approaches (sonnet for implementation)
+Task(subagent_type="general-purpose", prompt="Approach A...", model="sonnet", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="Approach B...", model="sonnet", run_in_background=True)
 
 # Notifications arrive → Evaluate and continue with winner
 winner = evaluate(approach_a, approach_b)
-Task(subagent_type="general-purpose", prompt=f"Refine and complete: {winner}", run_in_background=True)
+Task(subagent_type="general-purpose", prompt=f"Refine and complete: {winner}", model="sonnet", run_in_background=True)
 ```
 
 ---
@@ -307,7 +307,7 @@ if result.failed or result.incomplete:
         "content": f"Attempt 1 failed: {result.error}. Retrying with adjusted approach."
     })
 
-    # Retry with more context (still background)
+    # Retry with more context (still background, same model as original)
     Task(subagent_type="general-purpose",
          prompt=f"""Previous attempt failed: {result.error}
 
@@ -315,6 +315,7 @@ if result.failed or result.incomplete:
          - [specific guidance based on failure]
 
          Original task: [task description]""",
+         model="sonnet",
          run_in_background=True)
 ```
 
@@ -375,9 +376,9 @@ for result in [result1, result2, result3]:
 "Found these relevant files: {all_files}"
 ```
 
-**Complex synthesis (spawn synthesis agent):**
+**Complex synthesis (spawn synthesis agent - opus for judgment):**
 
-```python
+```
 Task(subagent_type="general-purpose",
      prompt=f"""Synthesize these parallel review findings into a unified report:
 
@@ -398,6 +399,7 @@ Create a single PR review with:
 - Positive notes
 
 Prioritize by severity. Remove duplicates. Do not mention that multiple reviews were conducted.""",
+     model="opus",
      run_in_background=True)
 ```
 
@@ -413,7 +415,7 @@ if has_conflict(result1, result2):
      Approach A: {result1.summary}
      Approach B: {result2.summary}"
 
-    # Option 2: Spawn resolution agent
+    # Option 2: Spawn resolution agent (opus for judgment)
     Task(subagent_type="general-purpose",
          prompt=f"""Two agents produced conflicting changes:
 
@@ -422,6 +424,7 @@ if has_conflict(result1, result2):
 
          Merge these changes, resolving conflicts by [priority rule].
          Ensure the final result is consistent.""",
+         model="opus",
          run_in_background=True)
 ```
 

package/dist/skills/orchestration/references/tools.md

@@ -201,10 +201,10 @@ Spawn an agent to handle work. This is how you delegate.
 
 **ALWAYS use `run_in_background=True`.** This is the foundation of powerful orchestration.
 
-```python
-# Correct: Background agents (ALWAYS)
-Task(subagent_type="Explore", prompt="...", run_in_background=True)
-Task(subagent_type="general-purpose", prompt="...", run_in_background=True)
+```
+# Correct: Background agents (ALWAYS) with explicit model
+Task(subagent_type="Explore", prompt="...", model="haiku", run_in_background=True)
+Task(subagent_type="general-purpose", prompt="...", model="sonnet", run_in_background=True)
 ```
 
 ### The Notification System
@@ -252,23 +252,23 @@ TaskOutput(task_id="abc123")
 
 ### Model Selection
 
-| Task Complexity        | Model           | Why                          |
-| ---------------------- | --------------- | ---------------------------- |
-| Simple search/patterns | `haiku`         | Fast and cheap               |
-| Standard exploration   | `haiku`         | Sufficient for most searches |
-| Complex exploration    | `sonnet`        | Needs reasoning              |
-| Simple implementation  | `haiku`         | Pattern-based work           |
-| Complex implementation | `sonnet`        | Design decisions needed      |
-| Architecture/planning  | `sonnet`/`opus` | Complex trade-offs           |
-| Security review        | `sonnet`        | Careful analysis             |
+| Task Type                        | Model    | Why                                      |
+| -------------------------------- | -------- | ---------------------------------------- |
+| Fetch files, grep, find things   | `haiku`  | Errand runner - spawn many in parallel   |
+| Gather info for synthesis        | `haiku`  | No judgment needed, just retrieval       |
+| Well-structured implementation   | `sonnet` | Capable worker - needs clear direction   |
+| Research, reading docs           | `sonnet` | Can follow patterns and instructions     |
+| Security review                  | `opus`   | Critical thinking, trust its judgment    |
+| Architecture/design decisions    | `opus`   | Ambiguous, needs creative problem-solving|
+| Complex debugging                | `opus`   | Reasoning across systems                 |
 
 ### Parallelism Strategy
 
-| Priority     | Approach                                         |
-| ------------ | ------------------------------------------------ |
-| **Speed**    | Parallelize with sonnet, accept higher cost      |
-| **Cost**     | Sequential haiku where possible                  |
-| **Balanced** | Haiku for exploration, sonnet for implementation |
+| Priority     | Approach                                           |
+| ------------ | -------------------------------------------------- |
+| **Speed**    | Swarm of haiku for gathering, parallel sonnet work |
+| **Cost**     | Haiku wherever possible, sonnet only when needed   |
+| **Balanced** | Haiku to gather, sonnet to implement, opus to decide |
 
 ---
 
@@ -482,10 +482,11 @@ TaskUpdate(taskId="3", addBlockedBy=["2"])
 # 3. Find ready (task 1 is unblocked)
 TaskList()
 
-# 4. Spawn background agent (ALWAYS background)
+# 4. Spawn background agent (ALWAYS background, explicit model)
 Task(subagent_type="general-purpose",
      description="Setup database",
      prompt="Create SQLite database with users table...",
+     model="sonnet",
      run_in_background=True)
 
 # 5. Update user and yield (or continue other work)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "cc-mirror",
-  "version": "1.4.0",
+  "version": "1.4.1",
   "type": "module",
   "description": "Create multiple isolated Claude Code variants with custom providers (Z.ai, MiniMax, OpenRouter, Claude Code Router)",
   "author": "Numman Ali",