code-muse 0.0.1__py3-none-any.whl

code_muse/plugins/shell_safety/register_callbacks.py

@@ -0,0 +1,202 @@
+"""Callback registration for shell command safety checking.
+
+This module registers a callback that intercepts shell commands in yolo_mode
+and assesses their safety risk before execution.
+"""
+
+from typing import Any
+
+from code_muse.callbacks import register_callback
+from code_muse.config import (
+    get_global_model_name,
+    get_safety_permission_level,
+    get_yolo_mode,
+)
+from code_muse.messaging import emit_info
+from code_muse.plugins.shell_safety.command_cache import (
+    cache_assessment,
+    get_cached_assessment,
+)
+from code_muse.tools.command_runner import ShellSafetyAssessment
+
+# OAuth model prefixes - these models have their own safety mechanisms
+OAUTH_MODEL_PREFIXES = (
+    "claude-code-",  # Anthropic OAuth
+    "chatgpt-",  # OpenAI OAuth
+    "gemini-oauth",  # Google OAuth
+)
+
+
+def is_oauth_model(model_name: str | None) -> bool:
+    """Check if the model is an OAuth model that should skip safety checks.
+
+    OAuth models have their own built-in safety mechanisms, so we skip
+    the shell safety callback to avoid redundant checks and potential bugs.
+
+    Args:
+        model_name: The name of the current model
+
+    Returns:
+        True if the model is an OAuth model, False otherwise
+    """
+    if not model_name:
+        return False
+    return model_name.startswith(OAUTH_MODEL_PREFIXES)
+
+
+# Risk level hierarchy for numeric comparison
+# Lower numbers = safer commands, higher numbers = more dangerous
+# This mapping allows us to compare risk levels as integers
+RISK_LEVELS: dict[str, int] = {
+    "none": 0,
+    "low": 1,
+    "medium": 2,
+    "high": 3,
+    "critical": 4,
+}
+
+
+def compare_risk_levels(assessed_risk: str | None, threshold: str) -> bool:
+    """Compare assessed risk against threshold.
+
+    Args:
+        assessed_risk: The risk level from the agent (can be None)
+        threshold: The configured risk threshold
+
+    Returns:
+        True if the command should be blocked (risk exceeds threshold)
+        False if the command is acceptable
+    """
+    # If assessment failed (None), treat as high risk (fail-safe behavior)
+    if assessed_risk is None:
+        assessed_risk = "high"
+
+    # Convert risk levels to numeric values for comparison
+    assessed_level = RISK_LEVELS.get(assessed_risk, 4)  # Default to critical if unknown
+    threshold_level = RISK_LEVELS.get(threshold, 2)  # Default to medium if unknown
+
+    # Block if assessed risk is GREATER than threshold
+    # Note: Commands AT the threshold level are allowed (>, not >=)
+    return assessed_level > threshold_level
+
+
+async def shell_safety_callback(
+    context: Any, command: str, cwd: str | None = None, timeout: int = 60
+) -> dict[str, Any | None]:
+    """Callback to assess shell command safety before execution.
+
+    This callback is only active when yolo_mode is True. When yolo_mode is False,
+    the user manually reviews every command, so we don't need the agent.
+
+    Args:
+        context: The execution context
+        command: The shell command to execute
+        cwd: Optional working directory
+        timeout: Command timeout (unused here)
+
+    Returns:
+        None if command is safe to proceed
+        Dict with rejection info if command should be blocked
+    """
+    # Skip safety checks for OAuth models - they have their own safety mechanisms
+    current_model = get_global_model_name()
+    if is_oauth_model(current_model):
+        return None
+
+    # Only check safety in yolo_mode - otherwise user is reviewing manually
+    yolo_mode = get_yolo_mode()
+    if not yolo_mode:
+        return None
+
+    # Get configured risk threshold
+    threshold = get_safety_permission_level()
+
+    try:
+        # Check cache first (fast path - no LLM call)
+        cached = get_cached_assessment(command, cwd)
+
+        if cached:
+            # Got a cached result - check against threshold
+            if compare_risk_levels(cached.risk, threshold):
+                # Cached result says it's too risky
+                risk_display = cached.risk or "unknown"
+                concise_reason = cached.reasoning or "No reasoning provided"
+                error_msg = (
+                    f"🛑 Command blocked (risk {risk_display.upper()} > permission {threshold.upper()}).\n"
+                    f"Reason: {concise_reason}\n"
+                    f"Override: /set yolo_mode true or /set safety_permission_level {risk_display}"
+                )
+                emit_info(error_msg)
+                return {
+                    "blocked": True,
+                    "risk": cached.risk,
+                    "reasoning": cached.reasoning,
+                    "error_message": error_msg,
+                }
+            # Cached result is within threshold - allow silently
+            return None
+
+        # Cache miss - need LLM assessment
+        # Import here to avoid circular imports
+        from code_muse.plugins.shell_safety.agent_shell_safety import ShellSafetyAgent
+
+        # Create agent and assess command
+        agent = ShellSafetyAgent()
+
+        # Build the assessment prompt with optional cwd context
+        prompt = f"Assess this shell command:\n\nCommand: {command}"
+        if cwd:
+            prompt += f"\nWorking directory: {cwd}"
+
+        # Run async assessment with structured output type
+        result = await agent.run(prompt, output_type=ShellSafetyAssessment)
+        assessment = result.output
+
+        # Cache the result for future use, but only if it's not a fallback assessment
+        if not getattr(assessment, "is_fallback", False):
+            cache_assessment(command, cwd, assessment.risk, assessment.reasoning)
+
+        # Check if risk exceeds threshold (commands at threshold are allowed)
+        if compare_risk_levels(assessment.risk, threshold):
+            risk_display = assessment.risk or "unknown"
+            concise_reason = assessment.reasoning or "No reasoning provided"
+            error_msg = (
+                f"🛑 Command blocked (risk {risk_display.upper()} > permission {threshold.upper()}).\n"
+                f"Reason: {concise_reason}\n"
+                f"Override: /set yolo_mode true or /set safety_permission_level {risk_display}"
+            )
+            emit_info(error_msg)
+
+            # Return rejection info for the command runner
+            return {
+                "blocked": True,
+                "risk": assessment.risk,
+                "reasoning": assessment.reasoning,
+                "error_message": error_msg,
+            }
+
+        # Command is within acceptable risk threshold - remain silent
+        return None  # Allow command to proceed
+
+    except Exception as e:
+        # On any error, fail safe by blocking the command
+        error_msg = (
+            f"🛑 Command blocked (risk HIGH > permission {threshold.upper()}).\n"
+            f"Reason: Safety assessment error: {str(e)}\n"
+            f"Override: /set yolo_mode true or /set safety_permission_level high"
+        )
+        return {
+            "blocked": True,
+            "risk": "high",
+            "reasoning": f"Safety assessment error: {str(e)}",
+            "error_message": error_msg,
+        }
+
+
+def register():
+    """Register the shell safety callback."""
+    register_callback("run_shell_command", shell_safety_callback, priority=50)
+
+
+# Auto-register the callback when this module is imported
+register()

code_muse/plugins/synthetic_status/__init__.py

@@ -0,0 +1 @@
+"""Synthetic provider status plugin."""

code_muse/plugins/synthetic_status/register_callbacks.py

@@ -0,0 +1,128 @@
+"""Slash commands for Synthetic provider subscription status."""
+
+from rich.panel import Panel
+
+from code_muse.callbacks import register_callback
+from code_muse.messaging import emit_error, emit_info, emit_warning
+from code_muse.model_factory import get_api_key
+
+from .status_api import fetch_synthetic_quota, resolve_syn_api_key
+
+_PROVIDER_ENV_KEYS = {
+    "synthetic": "SYN_API_KEY",
+    "openai": "OPENAI_API_KEY",
+    "anthropic": "ANTHROPIC_API_KEY",
+    "gemini": "GEMINI_API_KEY",
+    "cerebras": "CEREBRAS_API_KEY",
+    "openrouter": "OPENROUTER_API_KEY",
+    "zai": "ZAI_API_KEY",
+    "azure_openai": "AZURE_OPENAI_API_KEY",
+}
+
+
+def _custom_help() -> list[tuple[str, str]]:
+    return [
+        ("synthetic-status", "Check Synthetic subscription quota and renewal time"),
+        ("provider", "Provider utilities (usage: /provider synthetic status)"),
+        (
+            "status",
+            "Show provider status when only Synthetic appears configured",
+        ),
+    ]
+
+
+def _format_amount(value: float) -> str:
+    rounded = round(value)
+    if abs(value - rounded) < 1e-9:
+        return str(int(rounded))
+    return f"{value:.2f}".rstrip("0").rstrip(".")
+
+
+def _render_synthetic_status_panel(
+    limit: float,
+    used: float,
+    renews_at_utc: str,
+) -> Panel:
+    remaining = max(limit - used, 0.0)
+    body = "\n".join(
+        [
+            "Provider      : Synthetic",
+            f"Requests used : {_format_amount(used)} / {_format_amount(limit)}",
+            f"Requests left : {_format_amount(remaining)}",
+            f"Renews at     : {renews_at_utc}",
+        ]
+    )
+    return Panel(body, border_style="cyan")
+
+
+def _handle_synthetic_status() -> None:
+    api_key = resolve_syn_api_key()
+    if not api_key:
+        emit_error("SYN_API_KEY is not configured. Set it with /set syn_api_key <key>.")
+        return
+
+    result = fetch_synthetic_quota(api_key=api_key)
+    if not result.ok or not result.quota:
+        emit_warning(result.error or "Failed to fetch Synthetic quota status.")
+        return
+
+    renews_at_str = result.quota.renews_at_utc.strftime("%Y-%m-%d %H:%M UTC")
+    panel = _render_synthetic_status_panel(
+        limit=result.quota.limit,
+        used=result.quota.requests_used,
+        renews_at_utc=renews_at_str,
+    )
+    emit_info(panel)
+
+
+def _is_synthetic_only_provider_configured() -> bool:
+    configured = {
+        provider
+        for provider, env_name in _PROVIDER_ENV_KEYS.items()
+        if get_api_key(env_name)
+    }
+    return configured == {"synthetic"}
+
+
+def _handle_provider_command(command: str) -> bool | None:
+    tokens = command.strip().split()
+    if len(tokens) < 2:
+        return None
+
+    provider = tokens[1].lower()
+    if provider != "synthetic":
+        return None
+
+    if len(tokens) == 3 and tokens[2].lower() == "status":
+        _handle_synthetic_status()
+        return True
+
+    emit_info("Usage: /provider synthetic status")
+    return True
+
+
+def _handle_custom_command(command: str, name: str) -> bool | None:
+    if not name:
+        return None
+
+    if name == "synthetic-status":
+        _handle_synthetic_status()
+        return True
+
+    if name == "provider":
+        return _handle_provider_command(command)
+
+    if name == "status":
+        if _is_synthetic_only_provider_configured():
+            _handle_synthetic_status()
+        else:
+            emit_warning(
+                "Multiple providers appear configured. Use /provider synthetic status."
+            )
+        return True
+
+    return None
+
+
+register_callback("custom_command_help", _custom_help)
+register_callback("custom_command", _handle_custom_command)

code_muse/plugins/synthetic_status/status_api.py

@@ -0,0 +1,145 @@
+"""HTTP helpers for Synthetic quota status."""
+
+from dataclasses import dataclass
+from datetime import UTC, datetime
+
+import httpx
+
+from code_muse.model_factory import get_api_key
+
+SYNTHETIC_QUOTAS_URL = "https://api.synthetic.new/v2/quotas"
+
+
+@dataclass
+class SyntheticQuota:
+    """Parsed Synthetic subscription quota values."""
+
+    limit: float
+    requests_used: float
+    renews_at_utc: datetime
+
+
+@dataclass
+class SyntheticQuotaResult:
+    """Result wrapper for Synthetic quota fetches."""
+
+    quota: SyntheticQuota | None = None
+    error: str | None = None
+
+    @property
+    def ok(self) -> bool:
+        return self.quota is not None and self.error is None
+
+
+def resolve_syn_api_key() -> str | None:
+    """Resolve SYN_API_KEY from config/environment."""
+    value = get_api_key("SYN_API_KEY")
+    if not value:
+        return None
+    value = value.strip()
+    return value or None
+
+
+def _parse_renews_at(value: object) -> datetime | None:
+    if not isinstance(value, str) or not value.strip():
+        return None
+
+    normalized = value.strip().replace("Z", "+00:00")
+    try:
+        parsed = datetime.fromisoformat(normalized)
+    except ValueError:
+        return None
+
+    if parsed.tzinfo is None:
+        parsed = parsed.replace(tzinfo=UTC)
+    return parsed.astimezone(UTC)
+
+
+def fetch_synthetic_quota(
+    api_key: str,
+    timeout_seconds: float = 15.0,
+) -> SyntheticQuotaResult:
+    """Fetch and validate Synthetic subscription quota status."""
+    if not api_key:
+        return SyntheticQuotaResult(error="SYN_API_KEY is not configured.")
+
+    headers = {
+        "Authorization": f"Bearer {api_key}",
+        "Accept": "application/json",
+    }
+
+    try:
+        response = httpx.get(
+            SYNTHETIC_QUOTAS_URL,
+            headers=headers,
+            timeout=timeout_seconds,
+        )
+    except httpx.TimeoutException:
+        return SyntheticQuotaResult(error="Synthetic API request timed out.")
+    except httpx.ConnectError:
+        return SyntheticQuotaResult(
+            error="Could not connect to the Synthetic API endpoint."
+        )
+    except httpx.HTTPError as exc:
+        return SyntheticQuotaResult(error=f"Synthetic API request failed: {exc}")
+
+    if response.status_code in (401, 403):
+        return SyntheticQuotaResult(
+            error="Synthetic API authentication failed. Check SYN_API_KEY."
+        )
+    if response.status_code == 429:
+        return SyntheticQuotaResult(
+            error="Synthetic API rate limited this request (HTTP 429)."
+        )
+    if response.status_code >= 500:
+        return SyntheticQuotaResult(
+            error=f"Synthetic API server error (HTTP {response.status_code})."
+        )
+    if response.status_code != 200:
+        detail = response.text.strip()
+        if len(detail) > 200:
+            detail = f"{detail[:200]}..."
+        suffix = f": {detail}" if detail else ""
+        return SyntheticQuotaResult(
+            error=f"Synthetic API returned HTTP {response.status_code}{suffix}"
+        )
+
+    try:
+        payload = response.json()
+    except ValueError:
+        return SyntheticQuotaResult(
+            error="Synthetic API returned invalid JSON for /v2/quotas."
+        )
+
+    if not isinstance(payload, dict):
+        return SyntheticQuotaResult(
+            error="Synthetic API response payload is not an object."
+        )
+
+    subscription = payload.get("subscription")
+    if not isinstance(subscription, dict):
+        return SyntheticQuotaResult(
+            error="Synthetic API response is missing 'subscription'."
+        )
+
+    try:
+        limit = float(subscription.get("limit"))
+        requests_used = float(subscription.get("requests"))
+    except TypeError, ValueError:
+        return SyntheticQuotaResult(
+            error="Synthetic API response has invalid numeric quota values."
+        )
+
+    renews_at = _parse_renews_at(subscription.get("renewsAt"))
+    if renews_at is None:
+        return SyntheticQuotaResult(
+            error="Synthetic API response has an invalid 'renewsAt' timestamp."
+        )
+
+    return SyntheticQuotaResult(
+        quota=SyntheticQuota(
+            limit=limit,
+            requests_used=requests_used,
+            renews_at_utc=renews_at,
+        )
+    )

code_muse/plugins/token_caching/__init__.py

@@ -0,0 +1,21 @@
+"""Token caching plugin for Muse."""
+
+from code_muse.plugins.token_caching.cache_hit_tracking import (
+    CacheUsage,
+    SessionCacheStats,
+    _session_stats,
+    extract_cache_usage,
+)
+from code_muse.plugins.token_caching.cacheable_prefix_detection import (
+    detect_cache_breakpoint,
+)
+from code_muse.plugins.token_caching.stats_display import format_cache_stats
+
+__all__ = [
+    "CacheUsage",
+    "SessionCacheStats",
+    "_session_stats",
+    "detect_cache_breakpoint",
+    "extract_cache_usage",
+    "format_cache_stats",
+]

code_muse/plugins/token_caching/cache_hit_tracking.py

@@ -0,0 +1,128 @@
+"""Track Anthropic cache usage from API responses."""
+
+import threading
+from dataclasses import dataclass, field
+
+
+@dataclass
+class CacheUsage:
+    """Per-turn cache and token usage from an Anthropic API response."""
+
+    cache_read_tokens: int = 0
+    cache_write_tokens: int = 0
+    input_tokens: int = 0
+    output_tokens: int = 0
+
+
+@dataclass
+class SessionCacheStats:
+    """Thread-safe accumulator for cache usage across a session."""
+
+    # FREE-THREADED: Sync-only stats accumulator; keep threading.Lock.
+    _lock: threading.Lock = field(default_factory=threading.Lock, repr=False)
+    _total_read_tokens: int = 0
+    _total_write_tokens: int = 0
+    _total_input_tokens: int = 0
+    _total_output_tokens: int = 0
+
+    def record_usage(self, usage: CacheUsage) -> None:
+        """Accumulate a single turn's usage."""
+        with self._lock:
+            self._total_read_tokens += usage.cache_read_tokens
+            self._total_write_tokens += usage.cache_write_tokens
+            self._total_input_tokens += usage.input_tokens
+            self._total_output_tokens += usage.output_tokens
+
+    @property
+    def total_read_tokens(self) -> int:
+        with self._lock:
+            return self._total_read_tokens
+
+    @property
+    def total_write_tokens(self) -> int:
+        with self._lock:
+            return self._total_write_tokens
+
+    @property
+    def total_input_tokens(self) -> int:
+        with self._lock:
+            return self._total_input_tokens
+
+    @property
+    def total_output_tokens(self) -> int:
+        with self._lock:
+            return self._total_output_tokens
+
+    @property
+    def hit_rate(self) -> float:
+        """Cache hit rate: read tokens / (read tokens + input tokens)."""
+        with self._lock:
+            denominator = self._total_read_tokens + self._total_input_tokens
+            if denominator == 0:
+                return 0.0
+            return self._total_read_tokens / denominator
+
+    @property
+    def estimated_savings_usd(self) -> float:
+        """Estimated cost savings from token caching.
+
+        Anthropic pricing (Claude 3.5 Sonnet base):
+        - $3 / M input tokens  → base price per token = 3 / 1_000_000
+        - Cache read  = 0.1 × base
+        - Cache write = 1.25 × base
+
+        Savings = (read_tokens × 0.1 × base) - (write_tokens × 1.25 × base)
+        """
+        with self._lock:
+            base_price_per_token = 3.0 / 1_000_000.0
+            savings = (self._total_read_tokens * 0.1 * base_price_per_token) - (
+                self._total_write_tokens * 1.25 * base_price_per_token
+            )
+            return max(0.0, savings)
+
+    def reset(self) -> None:
+        """Clear all accumulated counters."""
+        with self._lock:
+            self._total_read_tokens = 0
+            self._total_write_tokens = 0
+            self._total_input_tokens = 0
+            self._total_output_tokens = 0
+
+
+def extract_cache_usage(response: dict) -> CacheUsage | None:
+    """Parse Anthropic API response dict for cache usage fields.
+
+    Looks for:
+    - ``usage.cache_read_input_tokens``
+    - ``usage.cache_creation_input_tokens``
+    - ``usage.input_tokens``
+    - ``usage.output_tokens``
+
+    Returns ``None`` if the response has no ``usage`` block.
+    Missing fields default to 0.
+    """
+    if not isinstance(response, dict):
+        return None
+
+    usage = response.get("usage")
+    if not isinstance(usage, dict):
+        return None
+
+    return CacheUsage(
+        cache_read_tokens=_int_or_zero(usage.get("cache_read_input_tokens")),
+        cache_write_tokens=_int_or_zero(usage.get("cache_creation_input_tokens")),
+        input_tokens=_int_or_zero(usage.get("input_tokens")),
+        output_tokens=_int_or_zero(usage.get("output_tokens")),
+    )
+
+
+def _int_or_zero(value: object) -> int:
+    """Coerce a value to int, defaulting to 0 on failure."""
+    try:
+        return int(value)  # type: ignore[arg-type]
+    except TypeError, ValueError:
+        return 0
+
+
+# Global singleton for the current session
+_session_stats: SessionCacheStats = SessionCacheStats()

code_muse/plugins/token_caching/cacheable_prefix_detection.py

@@ -0,0 +1,28 @@
+"""Detect the boundary between static (cacheable) prefix and dynamic suffix."""
+
+
+def detect_cache_breakpoint(messages: list[dict]) -> int:
+    """Find the boundary between static prefix and dynamic suffix.
+
+    Heuristic: everything before the first user message is static
+    (system prompt, project context, etc.). Returns the index of the
+    last static message (the breakpoint).
+
+    Args:
+        messages: Anthropic-style message list with ``role`` keys.
+
+    Returns:
+        Index of the last static message, or 0 if there is no static
+        prefix, or ``len(messages) - 1`` if every message is static.
+    """
+    if not messages:
+        return 0
+
+    for idx, msg in enumerate(messages):
+        if isinstance(msg, dict) and msg.get("role") == "user":
+            # First user message found — everything before it is static.
+            # The breakpoint is the message just before this one.
+            return max(0, idx - 1)
+
+    # No user message found — treat everything as static.
+    return len(messages) - 1