sky-dev 0.1.3__tar.gz → 0.1.4__tar.gz

{sky_dev-0.1.3 → sky_dev-0.1.4}/CHANGELOG.md

@@ -1,4 +1,28 @@
 # Changelog
+
+## [0.1.4] - 2026-09-07
+
+### Added
+- Config-driven role mapping in `sky.yaml` (`mode_roles`)
+- Circuit breaker for infinite tool-calling loops
+- 60-second timeout for agent responses
+- Security guardrails for command injection prevention
+
+### Changed
+- `planning` and `reviewer` roles now use `meta/muse-glimmer-30b` via NVIDIA NIM
+- Better error messages for 404 and 500 errors
+- Approval gate UI now shows `[y/n/e]` clearly
+
+### Fixed
+- `sky plan` now correctly uses `planning` role (was using `fast_loop`)
+- `sky ask` now correctly uses `general` role (was using `fast_loop`)
+- Non-existent model IDs replaced with verified models
+- Hardcoded `role = "fast_loop"` removed from `fast_loop.py`
+
+### Security
+- Command injection prevention (`[;&|`]` characters blocked)
+- Path traversal prevention
+- Better error messages for security violations
 ## [0.1.3] - 2026-09-07
 
 ### Added

{sky_dev-0.1.3 → sky_dev-0.1.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sky-dev
-Version: 0.1.3
+Version: 0.1.4
 Summary: Sky - Agentic coding assistant. Build without boundaries.
 Author-email: Aaditya A <aaditya@corover.ai>
 License: MIT
@@ -109,12 +109,19 @@ sky chat
 - 💰 **Cost Tracking** — Every session shows what it cost
 - 🔄 **Resumable Workflows** — Interrupt and resume anytime
 
-## Providers
+## Models
 
-| Provider | Use Case | Setup |
-|----------|----------|-------|
-| Groq | Planning, Routing, General | `GROQ_API_KEY` in .env |
-| NVIDIA NIM | Coding, Testing, Subagents | `NVIDIA_NIM_API_KEY` in .env |
+Sky uses specialized models for each task:
+
+| Role | Model | Provider |
+|------|-------|----------|
+| General Chat | GPT-OSS 120B | Groq |
+| Planning | Muse Glimmer 30B | NVIDIA NIM |
+| Review | Muse Glimmer 30B | NVIDIA NIM |
+| Routing | Compound Mini | Groq |
+| Tool Calling | Nemotron 120B | NVIDIA NIM |
+| Coding | Nemotron 120B | NVIDIA NIM |
+| Testing | Nemotron 120B | NVIDIA NIM |
 
 ## Commands
 

{sky_dev-0.1.3 → sky_dev-0.1.4}/README.md

@@ -60,12 +60,19 @@ sky chat
 - 💰 **Cost Tracking** — Every session shows what it cost
 - 🔄 **Resumable Workflows** — Interrupt and resume anytime
 
-## Providers
-
-| Provider | Use Case | Setup |
-|----------|----------|-------|
-| Groq | Planning, Routing, General | `GROQ_API_KEY` in .env |
-| NVIDIA NIM | Coding, Testing, Subagents | `NVIDIA_NIM_API_KEY` in .env |
+## Models
+
+Sky uses specialized models for each task:
+
+| Role | Model | Provider |
+|------|-------|----------|
+| General Chat | GPT-OSS 120B | Groq |
+| Planning | Muse Glimmer 30B | NVIDIA NIM |
+| Review | Muse Glimmer 30B | NVIDIA NIM |
+| Routing | Compound Mini | Groq |
+| Tool Calling | Nemotron 120B | NVIDIA NIM |
+| Coding | Nemotron 120B | NVIDIA NIM |
+| Testing | Nemotron 120B | NVIDIA NIM |
 
 ## Commands
 

{sky_dev-0.1.3 → sky_dev-0.1.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "sky-dev"
-version = "0.1.3"
+version = "0.1.4"
 description = "Sky - Agentic coding assistant. Build without boundaries."
 readme = "README.md"
 license = {text = "MIT"}

{sky_dev-0.1.3 → sky_dev-0.1.4}/sky/__init__.py

@@ -1,6 +1,6 @@
 """Sky - Build without boundaries."""
 
-__version__ = "0.1.3"
+__version__ = "0.1.4"
 __author__ = "Aaditya A"
 __author_email__ = "aaditya@corover.ai"
 __description__ = "Local, CLI-based, agentic software development assistant"

{sky_dev-0.1.3 → sky_dev-0.1.4}/sky/cli.py

@@ -739,18 +739,18 @@ def _write_default_models_yaml(path: Path):
         description: "Ultra-fast routing (0.1s)"
       - id: "openai/gpt-oss-120b"
         description: "Best general conversation"
-      - id: "meta-models/Muse-Glimmer-30B"
-        description: "Dedicated reasoning & planning"
       - id: "qwen/qwen3.6-27b"
         description: "Best-in-class tool calling"
 
   nim:
     base_url: "https://integrate.api.nvidia.com/v1"
     timeout: 60
-    default_model: "nvidia/nemotron-3-super-120b-a12b"
+    default_model: "meta/muse-glimmer-30b"
     models:
+      - id: "meta/muse-glimmer-30b"
+        description: "Purpose-built for agentic reasoning & planning"
       - id: "nvidia/nemotron-3-super-120b-a12b"
-        description: "Purpose-built for agentic workflows and tool calling"
+        description: "Best-in-class tool calling & coding"
       - id: "nvidia/llama-3.1-nemotron-70b-instruct"
         description: "Reliable backup model"
 
@@ -759,43 +759,43 @@ roles:
     provider: "groq"
     model_id: "openai/gpt-oss-120b"
     temperature: 0.7
-    description: "User interaction, chat, explanations"
+    description: "Best general conversation"
 
   planning:
-    provider: "groq"
-    model_id: "meta-models/Muse-Glimmer-30B"
+    provider: "nim"
+    model_id: "meta/muse-glimmer-30b"
     temperature: 0.3
-    description: "Task decomposition, structured planning"
+    description: "Dedicated reasoning & planning"
 
   reviewer:
-    provider: "groq"
-    model_id: "meta-models/Muse-Glimmer-30B"
+    provider: "nim"
+    model_id: "meta/muse-glimmer-30b"
     temperature: 0.3
-    description: "Code review, quality analysis"
+    description: "Code review & analysis"
 
   routing:
     provider: "groq"
     model_id: "groq/compound-mini"
     temperature: 0.0
-    description: "Intent classification, simple decisions"
+    description: "Ultra-fast routing (0.1s)"
 
   fast_loop:
     provider: "nim"
     model_id: "nvidia/nemotron-3-super-120b-a12b"
     temperature: 0.1
-    description: "Parallel tool calling, function execution"
+    description: "Best-in-class tool calling"
 
   coder:
     provider: "nim"
     model_id: "nvidia/nemotron-3-super-120b-a12b"
     temperature: 0.1
-    description: "Agentic coding, code generation"
+    description: "Agentic coding"
 
   tester:
     provider: "nim"
     model_id: "nvidia/nemotron-3-super-120b-a12b"
     temperature: 0.1
-    description: "Test generation, pattern recognition"
+    description: "Test generation & pattern recognition"
 
 fallback:
   provider: "nim"
@@ -932,10 +932,10 @@ def check_providers(global_install: bool = typer.Option(False, "--global", help=
     
         console.print("\n[bold]Model Recommendations:[/bold]")
         console.print("  - [bold]General Interaction:[/bold] openai/gpt-oss-120b (Groq) - Best conversation")
-        console.print("  - [bold]Planning/Reviewing:[/bold] Muse Glimmer (Groq) - Best reasoning")
+        console.print("  - [bold]Planning/Reviewing:[/bold] Muse Glimmer 30B (NIM) - Best reasoning")
         console.print("  - [bold]Routing:[/bold] groq/compound-mini (Groq) - Fastest (0.1s)")
-        console.print("  - [bold]Tool Calling:[/bold] Qwen 27b (Groq) - Best tool use")
-        console.print("  - [bold]Coding/Testing:[/bold] Devstral 2 (NIM) - Best SWE-bench (77.6%)")
+        console.print("  - [bold]Tool Calling:[/bold] Nemotron 120B (NIM) - Best tool use")
+        console.print("  - [bold]Coding/Testing:[/bold] Nemotron 120B (NIM) - Best SWE-bench")
     except SkyError as e:
         console.print(f"\n[bold red]Error ({e.code}):[/bold red] {e.message}")
         if e.suggestion:
@@ -969,3 +969,7 @@ if __name__ == "__main__":
     except Exception as e:
         console.print(f"[bold red]{_handle_error(e)}[/bold red]")
         sys.exit(1)
+        app()
+    except Exception as e:
+        console.print(f"[bold red]{_handle_error(e)}[/bold red]")
+        sys.exit(1)

{sky_dev-0.1.3 → sky_dev-0.1.4}/sky/config/schema.py

@@ -112,6 +112,15 @@ class DexProjectConfig(BaseModel):
     model_config = ConfigDict(extra="forbid")
     
     project_name: str = "sky-project"
+    mode_roles: Dict[str, str] = Field(
+        default={
+            "ask": "general",
+            "plan": "planning",
+            "agent": "fast_loop",
+            "workflow": "fast_loop",
+        },
+        description="Mapping from mode names to role names"
+    )
     approval_rules: List[ApprovalRuleConfig] = Field(default_factory=list)
     max_retries: int = Field(default=3, ge=1, le=10)
     docker_image: str = "python:3.11-slim"
@@ -295,7 +304,6 @@ def load_models_config(global_mode: bool = False) -> ModelRoutingConfig:
                 models=[
                     ModelInfoConfig(id="groq/compound-mini", description="Ultra-fast routing (0.1s)", context_window=8192, best_for=["routing", "classification"], provider="groq"),
                     ModelInfoConfig(id="openai/gpt-oss-120b", description="Best general conversation", context_window=128000, best_for=["general", "chat"], provider="groq"),
-                    ModelInfoConfig(id="meta-models/Muse-Glimmer-30B", description="Dedicated reasoning & planning", context_window=32768, best_for=["planning", "reviewing"], provider="groq"),
                     ModelInfoConfig(id="qwen/qwen3.6-27b", description="Best-in-class tool calling", context_window=32768, best_for=["tool_calling", "execution"], provider="groq")
                 ]
             ),
@@ -304,21 +312,22 @@ def load_models_config(global_mode: bool = False) -> ModelRoutingConfig:
                 timeout=60,
                 requires_api_key=True,
                 free_tier=True,
-                default_model="mistralai/devstral-2",
+                default_model="meta/muse-glimmer-30b",
                 models=[
-                    ModelInfoConfig(id="mistralai/devstral-2", description="Purpose-built for agentic coding", context_window=131072, best_for=["coding", "tool_use"], provider="nim"),
+                    ModelInfoConfig(id="meta/muse-glimmer-30b", description="Purpose-built for agentic reasoning & planning", context_window=131072, best_for=["planning", "reviewing"], provider="nim"),
+                    ModelInfoConfig(id="nvidia/nemotron-3-super-120b-a12b", description="Best-in-class tool calling & coding", context_window=131072, best_for=["coding", "tool_use"], provider="nim"),
                     ModelInfoConfig(id="nvidia/llama-3.1-nemotron-70b-instruct", description="Reliable backup model", context_window=131072, best_for=["fallback"], provider="nim")
                 ]
             )
         },
         roles={
             "general": ModelAssignmentConfig(provider="groq", model_id="openai/gpt-oss-120b", temperature=0.7),
-            "planning": ModelAssignmentConfig(provider="groq", model_id="meta-models/Muse-Glimmer-30B", temperature=0.3),
-            "reviewer": ModelAssignmentConfig(provider="groq", model_id="meta-models/Muse-Glimmer-30B", temperature=0.3),
+            "planning": ModelAssignmentConfig(provider="nim", model_id="meta/muse-glimmer-30b", temperature=0.3),
+            "reviewer": ModelAssignmentConfig(provider="nim", model_id="meta/muse-glimmer-30b", temperature=0.3),
             "routing": ModelAssignmentConfig(provider="groq", model_id="groq/compound-mini", temperature=0.0),
-            "fast_loop": ModelAssignmentConfig(provider="groq", model_id="qwen/qwen3.6-27b", temperature=0.1),
-            "coder": ModelAssignmentConfig(provider="nim", model_id="mistralai/devstral-2", temperature=0.1),
-            "tester": ModelAssignmentConfig(provider="nim", model_id="mistralai/devstral-2", temperature=0.1),
+            "fast_loop": ModelAssignmentConfig(provider="nim", model_id="nvidia/nemotron-3-super-120b-a12b", temperature=0.1),
+            "coder": ModelAssignmentConfig(provider="nim", model_id="nvidia/nemotron-3-super-120b-a12b", temperature=0.1),
+            "tester": ModelAssignmentConfig(provider="nim", model_id="nvidia/nemotron-3-super-120b-a12b", temperature=0.1),
         },
         fallback=ModelAssignmentConfig(provider="nim", model_id="nvidia/llama-3.1-nemotron-70b-instruct", temperature=0.1),
         timeout_seconds=30,

{sky_dev-0.1.3 → sky_dev-0.1.4}/sky/core/fast_loop.py

@@ -283,7 +283,10 @@ class FastLoopEngine:
             messages[0]["content"] = prompt
             
         tools = self._get_available_tools(mode, tools_filter)
-        role = "fast_loop"
+        
+        # Get mode-to-role mapping from config
+        mode_roles = getattr(self.config, "mode_roles", {})
+        role = mode_roles.get(mode, "fast_loop")
         
         if self.config.verbose:
             from rich.console import Console

{sky_dev-0.1.3 → sky_dev-0.1.4}/sky_dev.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sky-dev
-Version: 0.1.3
+Version: 0.1.4
 Summary: Sky - Agentic coding assistant. Build without boundaries.
 Author-email: Aaditya A <aaditya@corover.ai>
 License: MIT
@@ -109,12 +109,19 @@ sky chat
 - 💰 **Cost Tracking** — Every session shows what it cost
 - 🔄 **Resumable Workflows** — Interrupt and resume anytime
 
-## Providers
+## Models
 
-| Provider | Use Case | Setup |
-|----------|----------|-------|
-| Groq | Planning, Routing, General | `GROQ_API_KEY` in .env |
-| NVIDIA NIM | Coding, Testing, Subagents | `NVIDIA_NIM_API_KEY` in .env |
+Sky uses specialized models for each task:
+
+| Role | Model | Provider |
+|------|-------|----------|
+| General Chat | GPT-OSS 120B | Groq |
+| Planning | Muse Glimmer 30B | NVIDIA NIM |
+| Review | Muse Glimmer 30B | NVIDIA NIM |
+| Routing | Compound Mini | Groq |
+| Tool Calling | Nemotron 120B | NVIDIA NIM |
+| Coding | Nemotron 120B | NVIDIA NIM |
+| Testing | Nemotron 120B | NVIDIA NIM |
 
 ## Commands