htmlgraph 0.9.3__py3-none-any.whl → 0.27.5__py3-none-any.whl

htmlgraph/orchestration/spawners/gemini.py

@@ -0,0 +1,471 @@
+"""Gemini spawner implementation."""
+
+import json
+import logging
+import subprocess
+import time
+from typing import TYPE_CHECKING, Any
+
+from .base import AIResult, BaseSpawner
+
+logger = logging.getLogger(__name__)
+
+if TYPE_CHECKING:
+    from htmlgraph.sdk import SDK
+
+
+class GeminiSpawner(BaseSpawner):
+    """Spawner for Google Gemini CLI.
+
+    Model Selection:
+        The `model` parameter defaults to None, which is the RECOMMENDED approach.
+        When model=None, the Gemini CLI automatically selects the best available model
+        based on the task and current availability.
+
+        As of Gemini CLI v0.22+, the default models include:
+        - gemini-2.5-flash-lite: Fast, efficient model for most tasks
+        - gemini-3-flash-preview: Preview of Gemini 3 with enhanced capabilities
+
+        Explicitly specifying a model is DISCOURAGED because:
+        1. Older models (gemini-2.0-flash, gemini-1.5-flash) may fail due to
+           "thinking mode" incompatibility in newer CLI versions
+        2. Using None automatically benefits from Google's model updates
+        3. The CLI handles model selection and fallback logic
+
+        Supported models (if you must specify):
+        - None (recommended): CLI chooses best available model
+        - "gemini-2.5-flash-lite": Fast, efficient
+        - "gemini-3-flash-preview": Gemini 3 preview (enhanced capabilities)
+        - "gemini-2.5-pro": More capable, slower
+
+        DEPRECATED models (may cause errors):
+        - "gemini-2.0-flash": Deprecated, use None instead
+        - "gemini-1.5-flash": Deprecated, use None instead
+        - "gemini-1.5-pro": Deprecated, use None instead
+    """
+
+    def _parse_and_track_events(self, jsonl_output: str, sdk: "SDK") -> list[dict]:
+        """
+        Parse Gemini stream-json events and track in HtmlGraph.
+
+        Args:
+            jsonl_output: JSONL output from Gemini CLI
+            sdk: HtmlGraph SDK instance for tracking
+
+        Returns:
+            Parsed events list
+        """
+        events = []
+
+        for line in jsonl_output.splitlines():
+            if not line.strip():
+                continue
+
+            try:
+                event = json.loads(line)
+                events.append(event)
+
+                # Track based on event type
+                event_type = event.get("type")
+
+                if event_type == "tool_use":
+                    tool_name = event.get("tool_name", "unknown_tool")
+                    parameters = event.get("parameters", {})
+                    self._track_activity(
+                        sdk,
+                        tool="gemini_tool_call",
+                        summary=f"Gemini called {tool_name}",
+                        payload={
+                            "tool_name": tool_name,
+                            "parameters": parameters,
+                        },
+                    )
+
+                elif event_type == "tool_result":
+                    status = event.get("status", "unknown")
+                    success = status == "success"
+                    tool_id = event.get("tool_id", "unknown")
+                    self._track_activity(
+                        sdk,
+                        tool="gemini_tool_result",
+                        summary=f"Gemini tool result: {status}",
+                        success=success,
+                        payload={"tool_id": tool_id, "status": status},
+                    )
+
+                elif event_type == "message":
+                    role = event.get("role")
+                    if role == "assistant":
+                        content = event.get("content", "")
+                        # Truncate for summary
+                        summary = (
+                            content[:100] + "..." if len(content) > 100 else content
+                        )
+                        self._track_activity(
+                            sdk,
+                            tool="gemini_message",
+                            summary=f"Gemini: {summary}",
+                            payload={"role": role, "content_length": len(content)},
+                        )
+
+                elif event_type == "result":
+                    stats = event.get("stats", {})
+                    self._track_activity(
+                        sdk,
+                        tool="gemini_completion",
+                        summary="Gemini task completed",
+                        payload={"stats": stats},
+                    )
+
+            except json.JSONDecodeError:
+                # Skip malformed lines
+                continue
+
+        return events
+
+    def spawn(
+        self,
+        prompt: str,
+        output_format: str = "stream-json",
+        model: str | None = None,
+        include_directories: list[str] | None = None,
+        track_in_htmlgraph: bool = True,
+        timeout: int = 120,
+        tracker: Any = None,
+        parent_event_id: str | None = None,
+    ) -> AIResult:
+        """
+        Spawn Gemini in headless mode.
+
+        Args:
+            prompt: Task description for Gemini
+            output_format: "json" or "stream-json" (enables real-time tracking)
+            model: Model selection. Default: None (RECOMMENDED).
+
+                   When model=None (default), the Gemini CLI automatically selects
+                   the best available model, which includes:
+                   - gemini-2.5-flash-lite: Fast, efficient model
+                   - gemini-3-flash-preview: Gemini 3 with enhanced capabilities
+
+                   Using None is STRONGLY RECOMMENDED because:
+                   1. Automatically benefits from Google's latest models
+                   2. Avoids deprecation issues with older model names
+                   3. CLI handles optimal model selection and fallback
+
+                   DEPRECATED models (may cause errors with CLI v0.22+):
+                   - gemini-2.0-flash, gemini-1.5-flash, gemini-1.5-pro
+
+            include_directories: Directories to include for context. Default: None
+            track_in_htmlgraph: Enable HtmlGraph activity tracking. Default: True
+            timeout: Max seconds to wait
+            tracker: Optional SpawnerEventTracker for recording subprocess invocation
+            parent_event_id: Optional parent event ID for event hierarchy
+
+        Returns:
+            AIResult with response, error, and tracked events if tracking enabled
+
+        Example:
+            >>> spawner = GeminiSpawner()
+            >>> result = spawner.spawn(
+            ...     prompt="Analyze this codebase",
+            ...     # model=None is the default - uses latest Gemini models
+            ...     track_in_htmlgraph=True
+            ... )
+        """
+        # Initialize tracking if enabled
+        sdk: SDK | None = None
+        tracked_events: list[dict] = []
+        if track_in_htmlgraph:
+            sdk = self._get_sdk()
+
+        # Publish live event: spawner starting
+        self._publish_live_event(
+            "spawner_start",
+            "gemini",
+            prompt=prompt,
+            model=model,
+        )
+        start_time = time.time()
+
+        try:
+            # Build command based on tested pattern from spike spk-4029eef3
+            cmd = ["gemini", "-p", prompt, "--output-format", output_format]
+
+            # Add model option if specified
+            if model:
+                cmd.extend(["-m", model])
+
+            # Add include directories if specified
+            if include_directories:
+                for directory in include_directories:
+                    cmd.extend(["--include-directories", directory])
+
+            # CRITICAL: Add --yolo for headless mode (auto-approve all tools)
+            cmd.append("--yolo")
+
+            # Track spawner start if SDK available
+            if sdk:
+                self._track_activity(
+                    sdk,
+                    tool="gemini_spawn_start",
+                    summary=f"Spawning Gemini: {prompt[:80]}",
+                    payload={"prompt_length": len(prompt), "model": model},
+                )
+
+            # Publish live event: executing
+            self._publish_live_event(
+                "spawner_phase",
+                "gemini",
+                phase="executing",
+                details="Running Gemini CLI",
+            )
+
+            # Record subprocess invocation if tracker is available
+            subprocess_event_id = None
+            logger.warning(
+                f"DEBUG: tracker={tracker is not None}, parent_event_id={parent_event_id}"
+            )
+            if tracker and parent_event_id:
+                logger.debug("Recording subprocess invocation for Gemini...")
+                try:
+                    subprocess_event = tracker.record_tool_call(
+                        tool_name="subprocess.gemini",
+                        tool_input={"cmd": cmd},
+                        phase_event_id=parent_event_id,
+                        spawned_agent=model or "gemini-default",
+                    )
+                    if subprocess_event:
+                        subprocess_event_id = subprocess_event.get("event_id")
+                        logger.warning(
+                            f"DEBUG: Subprocess event created for Gemini: {subprocess_event_id}"
+                        )
+                    else:
+                        logger.debug("subprocess_event was None")
+                except Exception as e:
+                    # Tracking failure should not break execution
+                    logger.warning(f"DEBUG: Exception recording Gemini subprocess: {e}")
+                    pass
+            else:
+                logger.warning(
+                    f"DEBUG: Skipping Gemini subprocess tracking - tracker={tracker is not None}, parent_event_id={parent_event_id}"
+                )
+
+            # Execute with timeout and stderr redirection
+            # Note: Cannot use capture_output with stderr parameter
+            result = subprocess.run(
+                cmd,
+                stdout=subprocess.PIPE,
+                stderr=subprocess.DEVNULL,  # Redirect stderr to avoid polluting JSON
+                text=True,
+                timeout=timeout,
+            )
+
+            # Complete subprocess invocation tracking
+            if tracker and subprocess_event_id:
+                try:
+                    tracker.complete_tool_call(
+                        event_id=subprocess_event_id,
+                        output_summary=result.stdout[:500] if result.stdout else "",
+                        success=result.returncode == 0,
+                    )
+                except Exception:
+                    # Tracking failure should not break execution
+                    pass
+
+            # Publish live event: processing response
+            self._publish_live_event(
+                "spawner_phase",
+                "gemini",
+                phase="processing",
+                details="Parsing Gemini response",
+            )
+
+            # Check for command execution errors
+            if result.returncode != 0:
+                duration = time.time() - start_time
+                self._publish_live_event(
+                    "spawner_complete",
+                    "gemini",
+                    success=False,
+                    duration=duration,
+                    error=f"CLI failed with exit code {result.returncode}",
+                )
+                return AIResult(
+                    success=False,
+                    response="",
+                    tokens_used=None,
+                    error=f"Gemini CLI failed with exit code {result.returncode}",
+                    raw_output=None,
+                    tracked_events=tracked_events,
+                )
+
+            # Handle stream-json format with real-time tracking
+            if output_format == "stream-json" and sdk:
+                try:
+                    tracked_events = self._parse_and_track_events(result.stdout, sdk)
+                    # Only use stream-json parsing if we got valid events
+                    if tracked_events:
+                        # For stream-json, we need to extract response differently
+                        # Collect all assistant message content, then check result
+                        response_text = ""
+                        for event in tracked_events:
+                            if event.get("type") == "message":
+                                # Only collect assistant messages
+                                if event.get("role") == "assistant":
+                                    content = event.get("content", "")
+                                    if content:
+                                        response_text += content
+                            elif event.get("type") == "result":
+                                # Result event may have response field (override if present)
+                                if "response" in event and event["response"]:
+                                    response_text = event["response"]
+                                # Don't break - we've already collected messages
+
+                        # Token usage from stats in result event
+                        tokens = None
+                        for event in tracked_events:
+                            if event.get("type") == "result":
+                                stats = event.get("stats", {})
+                                if stats and "models" in stats:
+                                    total_tokens = 0
+                                    for model_stats in stats["models"].values():
+                                        model_tokens = model_stats.get(
+                                            "tokens", {}
+                                        ).get("total", 0)
+                                        total_tokens += model_tokens
+                                    tokens = total_tokens if total_tokens > 0 else None
+                                break
+
+                        # Publish live event: complete
+                        duration = time.time() - start_time
+                        self._publish_live_event(
+                            "spawner_complete",
+                            "gemini",
+                            success=True,
+                            duration=duration,
+                            response=response_text,
+                            tokens=tokens,
+                        )
+                        return AIResult(
+                            success=True,
+                            response=response_text,
+                            tokens_used=tokens,
+                            error=None,
+                            raw_output={"events": tracked_events},
+                            tracked_events=tracked_events,
+                        )
+
+                except Exception:
+                    # Fall back to regular JSON parsing if tracking fails
+                    pass
+
+            # Parse JSON response (for json format or fallback)
+            try:
+                output = json.loads(result.stdout)
+            except json.JSONDecodeError as e:
+                duration = time.time() - start_time
+                self._publish_live_event(
+                    "spawner_complete",
+                    "gemini",
+                    success=False,
+                    duration=duration,
+                    error=f"Failed to parse JSON: {e}",
+                )
+                return AIResult(
+                    success=False,
+                    response="",
+                    tokens_used=None,
+                    error=f"Failed to parse JSON output: {e}",
+                    raw_output={"stdout": result.stdout},
+                    tracked_events=tracked_events,
+                )
+
+            # Extract response and token usage from parsed output
+            # Response is at top level in JSON output
+            response_text = output.get("response", "")
+
+            # Token usage is in stats.models (sum across all models)
+            tokens = None
+            stats = output.get("stats", {})
+            if stats and "models" in stats:
+                total_tokens = 0
+                for model_stats in stats["models"].values():
+                    model_tokens = model_stats.get("tokens", {}).get("total", 0)
+                    total_tokens += model_tokens
+                tokens = total_tokens if total_tokens > 0 else None
+
+            # Publish live event: complete
+            duration = time.time() - start_time
+            self._publish_live_event(
+                "spawner_complete",
+                "gemini",
+                success=True,
+                duration=duration,
+                response=response_text,
+                tokens=tokens,
+            )
+            return AIResult(
+                success=True,
+                response=response_text,
+                tokens_used=tokens,
+                error=None,
+                raw_output=output,
+                tracked_events=tracked_events,
+            )
+
+        except subprocess.TimeoutExpired as e:
+            duration = time.time() - start_time
+            self._publish_live_event(
+                "spawner_complete",
+                "gemini",
+                success=False,
+                duration=duration,
+                error=f"Timed out after {timeout} seconds",
+            )
+            return AIResult(
+                success=False,
+                response="",
+                tokens_used=None,
+                error=f"Gemini CLI timed out after {timeout} seconds",
+                raw_output={
+                    "partial_stdout": e.stdout.decode() if e.stdout else None,
+                    "partial_stderr": e.stderr.decode() if e.stderr else None,
+                }
+                if e.stdout or e.stderr
+                else None,
+                tracked_events=tracked_events,
+            )
+        except FileNotFoundError:
+            duration = time.time() - start_time
+            self._publish_live_event(
+                "spawner_complete",
+                "gemini",
+                success=False,
+                duration=duration,
+                error="CLI not found",
+            )
+            return AIResult(
+                success=False,
+                response="",
+                tokens_used=None,
+                error="Gemini CLI not found. Ensure 'gemini' is installed and in PATH.",
+                raw_output=None,
+                tracked_events=tracked_events,
+            )
+        except Exception as e:
+            duration = time.time() - start_time
+            self._publish_live_event(
+                "spawner_complete",
+                "gemini",
+                success=False,
+                duration=duration,
+                error=str(e),
+            )
+            return AIResult(
+                success=False,
+                response="",
+                tokens_used=None,
+                error=f"Unexpected error: {type(e).__name__}: {e}",
+                raw_output=None,
+                tracked_events=tracked_events,
+            )

htmlgraph/orchestration/subprocess_runner.py

@@ -0,0 +1,36 @@
+from __future__ import annotations
+
+"""Subprocess execution with standardized error handling.
+
+Provides consistent error handling for Claude Code CLI invocations.
+"""
+
+import logging
+import subprocess
+import sys
+
+logger = logging.getLogger(__name__)
+
+
+class SubprocessRunner:
+    """Execute subprocess commands with error handling."""
+
+    @staticmethod
+    def run_claude_command(cmd: list[str]) -> None:
+        """Execute Claude Code CLI command with error handling.
+
+        Args:
+            cmd: Command list (e.g., ["claude", "--resume"])
+
+        Raises:
+            SystemExit: If 'claude' command not found or other error
+        """
+        try:
+            subprocess.run(cmd, check=False)
+        except FileNotFoundError:
+            logger.warning("Error: 'claude' command not found.")
+            print(
+                "Please install Claude Code CLI: https://code.claude.com",
+                file=sys.stderr,
+            )
+            sys.exit(1)