fix(nim): update model aliases to currently-available NIM models

- Remove dead anthropic/claude-* aliases (not on NIM)
- Add haiku, kimi, qwen, nemotron, gpt-oss aliases for real NIM models
- Update 6 default agents to use working NIM models:
  - coder: qwen3-coder-480b (code generation)
  - tester: nemotron-nano-8b (cheap)
  - reviewer: llama-3.3-70b (strong reasoning)
  - docs: nemotron-nano-8b (cheap, low temp)
  - security: kimi-k2.6 (good for code review)
  - architect: gpt-oss-120b (reasoning)
- Reduce default timeout from 5m to 2m

Verified with real nvapi key: orchestrator-run 'Add hello world' returns
success=true, 551 tokens, 0.0006 USD via NIM.

Author: Delqhi

cmd/sin-code/internal/llm/nim.go

@@ -1,23 +1,34 @@
 // SPDX-License-Identifier: MIT
 // Purpose: NVIDIA NIM-specific helpers. NIM exposes an OpenAI-compatible
 // chat/completions endpoint at https://integrate.api.nvidia.com/v1 and
-// serves Anthropic / Llama / Mistral / etc. models under "vendor/name"
-// IDs. Friendly aliases like "haiku" or "sonnet" resolve to the full ID.
+// serves many vendor models. Friendly aliases like "haiku" or "qwen"
+// resolve to a working NIM model ID.
 package llm
 
 const NIMDefaultBaseURL = "https://integrate.api.nvidia.com/v1"
 
-const NIMDefaultModel = "meta/llama-3.1-70b-instruct"
+const NIMDefaultModel = "meta/llama-3.3-70b-instruct"
 
-const NIMClaudeModel = "anthropic/claude-3-5-sonnet-20241022"
+const NIMHaikuModel = "nvidia/llama-3.1-nemotron-nano-8b-v1"
 
-const NIMHaikuModel = "anthropic/claude-3-5-haiku-20241022"
+const NIMKimiModel = "moonshotai/kimi-k2.6"
+
+const NIMQwenModel = "qwen/qwen3-coder-480b-a35b-instruct"
+
+const NIMNemotronModel = "nvidia/nemotron-3-nano-30b-a3b"
+
+const NIMGptOssModel = "openai/gpt-oss-120b"
 
 var NIMModelAliases = map[string]string{
-	"haiku":     NIMHaikuModel,
-	"sonnet":    NIMClaudeModel,
-	"llama-70b": NIMDefaultModel,
-	"llama-8b":  "meta/llama-3.1-8b-instruct",
+	"haiku":         NIMHaikuModel,
+	"kimi":          NIMKimiModel,
+	"qwen":          NIMQwenModel,
+	"nemotron":      NIMNemotronModel,
+	"gpt-oss":       NIMGptOssModel,
+	"llama-70b":     NIMDefaultModel,
+	"llama-3.3-70b": NIMDefaultModel,
+	"llama-8b":      "meta/llama-3.1-8b-instruct",
+	"default":       NIMDefaultModel,
 }
 
 func ResolveModel(name string) string {

cmd/sin-code/internal/orchestrator/agents.go

@@ -53,7 +53,7 @@ func DefaultAgents() []AgentConfig {
 			Name:        "coder",
 			Description: "Writes production-quality code following project conventions",
 			Type:        TaskCode,
-			Model:       "anthropic/claude-sonnet-4.7",
+			Model:       "qwen/qwen3-coder-480b-a35b-instruct",
 			MaxTokens:   16000,
 			Temperature: 0.0,
 			SystemFile:  "agents/coder/system.md",
@@ -67,7 +67,7 @@ func DefaultAgents() []AgentConfig {
 			Name:        "tester",
 			Description: "Writes unit, integration, and end-to-end tests",
 			Type:        TaskTest,
-			Model:       "anthropic/claude-haiku-4.5",
+			Model:       "nvidia/llama-3.1-nemotron-nano-8b-v1",
 			MaxTokens:   8000,
 			Temperature: 0.0,
 			SystemFile:  "agents/tester/system.md",
@@ -80,7 +80,7 @@ func DefaultAgents() []AgentConfig {
 			Name:        "reviewer",
 			Description: "Senior engineer reviewing code for correctness, style, and test coverage",
 			Type:        TaskReview,
-			Model:       "anthropic/claude-opus-5.1",
+			Model:       "meta/llama-3.3-70b-instruct",
 			MaxTokens:   8000,
 			Temperature: 0.0,
 			SystemFile:  "agents/reviewer/system.md",
@@ -93,7 +93,7 @@ func DefaultAgents() []AgentConfig {
 			Name:        "docs",
 			Description: "Technical writer producing clear, accurate documentation",
 			Type:        TaskDocs,
-			Model:       "anthropic/claude-haiku-4.5",
+			Model:       "nvidia/llama-3.1-nemotron-nano-8b-v1",
 			MaxTokens:   8000,
 			Temperature: 0.2,
 			SystemFile:  "agents/docs/system.md",
@@ -106,7 +106,7 @@ func DefaultAgents() []AgentConfig {
 			Name:        "security",
 			Description: "Application security specialist scanning for vulnerabilities",
 			Type:        TaskSecurity,
-			Model:       "anthropic/claude-sonnet-4.7",
+			Model:       "moonshotai/kimi-k2.6",
 			MaxTokens:   8000,
 			Temperature: 0.0,
 			SystemFile:  "agents/security/system.md",
@@ -119,7 +119,7 @@ func DefaultAgents() []AgentConfig {
 			Name:        "architect",
 			Description: "Principal architect designing high-level solutions",
 			Type:        TaskArchitect,
-			Model:       "anthropic/claude-opus-5.1",
+			Model:       "openai/gpt-oss-120b",
 			MaxTokens:   16000,
 			Temperature: 0.1,
 			SystemFile:  "agents/architect/system.md",

cmd/sin-code/internal/orchestrator_cmd.go

@@ -104,7 +104,7 @@ var OrchestratorPlanCmd = &cobra.Command{
 
 func init() {
 	OrchestratorRunCmd.Flags().StringVar(&orch2Format, "format", "text", "Output format: text|json")
-	OrchestratorRunCmd.Flags().DurationVar(&orch2Timeout, "timeout", 5*time.Minute, "Max execution time")
+	OrchestratorRunCmd.Flags().DurationVar(&orch2Timeout, "timeout", 2*time.Minute, "Max execution time")
 	OrchestratorRunCmd.Flags().IntVar(&orch2MaxParallel, "max-parallel", 4, "Max parallel agents")
 	OrchestratorRunCmd.Flags().StringVar(&orch2AgentsDir, "agents-dir", "", "User agents dir (default ~/.config/sin-code/agents)")
 	OrchestratorRunCmd.Flags().BoolVar(&orch2PlanOnly, "plan-only", false, "Build plan and exit, no execution")