foundry-mcp 0.3.3__py3-none-any.whl → 0.8.10__py3-none-any.whl

foundry_mcp/core/llm_config.py

@@ -163,6 +163,14 @@ class ProviderSpec:
             "[api]provider/model or [cli]transport[:backend/model|:model]"
         )
 
+    @classmethod
+    def parse_flexible(cls, spec: str) -> "ProviderSpec":
+        """Parse with fallback for simple provider IDs."""
+        spec = spec.strip()
+        if spec.startswith("["):
+            return cls.parse(spec)
+        return cls(type="cli", provider=spec.lower(), raw=spec)
+
     def validate(self) -> List[str]:
         """Validate the provider specification.
 
@@ -860,16 +868,18 @@ def reset_workflow_config() -> None:
 class WorkflowConsultationConfig:
     """Per-workflow consultation configuration overrides.
 
-    Allows individual workflows to specify minimum model requirements
-    and timeout overrides for AI consultations.
+    Allows individual workflows to specify minimum model requirements,
+    timeout overrides, and default review types for AI consultations.
 
     TOML Configuration Example:
         [consultation.workflows.fidelity_review]
         min_models = 2
         timeout_override = 600.0
+        default_review_type = "full"
 
         [consultation.workflows.plan_review]
         min_models = 3
+        default_review_type = "full"
 
     Attributes:
         min_models: Minimum number of models required for consensus (default: 1).
@@ -878,10 +888,17 @@ class WorkflowConsultationConfig:
         timeout_override: Optional timeout override in seconds. When set,
                           overrides the default_timeout from ConsultationConfig
                           for this specific workflow.
+        default_review_type: Default review type for this workflow (default: "full").
+                             Valid values: "quick", "full", "security", "feasibility".
+                             Used when no explicit review_type is provided in requests.
     """
 
     min_models: int = 1
     timeout_override: Optional[float] = None
+    default_review_type: str = "full"
+
+    # Valid review types
+    VALID_REVIEW_TYPES = {"quick", "full", "security", "feasibility"}
 
     def validate(self) -> None:
         """Validate the workflow consultation configuration.
@@ -897,6 +914,12 @@ class WorkflowConsultationConfig:
                 f"timeout_override must be positive if set, got {self.timeout_override}"
             )
 
+        if self.default_review_type not in self.VALID_REVIEW_TYPES:
+            raise ValueError(
+                f"default_review_type must be one of {sorted(self.VALID_REVIEW_TYPES)}, "
+                f"got '{self.default_review_type}'"
+            )
+
     @classmethod
     def from_dict(cls, data: Dict[str, Any]) -> "WorkflowConsultationConfig":
         """Create WorkflowConsultationConfig from a dictionary.
@@ -917,6 +940,9 @@ class WorkflowConsultationConfig:
             if value is not None:
                 config.timeout_override = float(value)
 
+        if "default_review_type" in data:
+            config.default_review_type = str(data["default_review_type"]).lower()
+
         return config
 
 

foundry_mcp/core/metrics_store.py

@@ -180,7 +180,7 @@ class FileMetricsStore(MetricsStore):
     for efficient querying. Thread-safe with file locking for concurrent access.
 
     Directory structure:
-        .cache/foundry-mcp/metrics/
+        ~/.foundry-mcp/metrics/
             metrics.jsonl    - Append-only metrics log
             index.json       - Metric name -> metadata mapping
     """
@@ -628,7 +628,7 @@ def get_metrics_store(storage_path: Optional[str | Path] = None) -> MetricsStore
         if _metrics_store is None:
             if storage_path is None:
                 # Default path
-                storage_path = Path.home() / ".cache" / "foundry-mcp" / "metrics"
+                storage_path = Path.home() / ".foundry-mcp" / "metrics"
             _metrics_store = FileMetricsStore(storage_path)
 
         return _metrics_store

foundry_mcp/core/naming.py

@@ -4,17 +4,34 @@ from __future__ import annotations
 
 import asyncio
 import functools
+import json
 import logging
 import time
 from typing import Any, Callable
 
 from mcp.server.fastmcp import FastMCP
+from mcp.types import TextContent
 
 from foundry_mcp.core.observability import mcp_tool
 
 logger = logging.getLogger(__name__)
 
 
+def _minify_response(result: dict[str, Any]) -> TextContent:
+    """Convert dict to TextContent with minified JSON.
+
+    Args:
+        result: Dictionary to serialize
+
+    Returns:
+        TextContent with minified JSON string
+    """
+    return TextContent(
+        type="text",
+        text=json.dumps(result, separators=(",", ":"), default=str),
+    )
+
+
 def canonical_tool(
     mcp: FastMCP,
     *,
@@ -45,7 +62,10 @@ def canonical_tool(
                 """Async wrapper for async underlying functions."""
                 start_time = time.perf_counter()
                 try:
-                    return await func(*args, **kwargs)
+                    result = await func(*args, **kwargs)
+                    if isinstance(result, dict):
+                        return _minify_response(result)
+                    return result
                 except Exception as e:
                     duration_ms = (time.perf_counter() - start_time) * 1000
                     _collect_tool_error(
@@ -64,7 +84,10 @@ def canonical_tool(
                 """Sync wrapper for sync underlying functions."""
                 start_time = time.perf_counter()
                 try:
-                    return func(*args, **kwargs)
+                    result = func(*args, **kwargs)
+                    if isinstance(result, dict):
+                        return _minify_response(result)
+                    return result
                 except Exception as e:
                     duration_ms = (time.perf_counter() - start_time) * 1000
                     _collect_tool_error(

foundry_mcp/core/progress.py

@@ -315,3 +315,73 @@ def get_task_counts_by_status(spec_data: Dict[str, Any]) -> Dict[str, int]:
                 counts[status] += 1
 
     return counts
+
+
+def sync_computed_fields(spec_data: Dict[str, Any]) -> Dict[str, Any]:
+    """
+    Synchronize computed fields to their canonical top-level locations.
+
+    This function should be called after any task status change to ensure
+    progress_percentage, current_phase, and status are persisted to the spec.
+
+    Updates (in-place):
+    - progress_percentage: calculated from hierarchy counts
+    - current_phase: first in_progress phase, or first pending if none
+    - status: based on overall progress
+
+    Args:
+        spec_data: Spec data dictionary (modified in place)
+
+    Returns:
+        Dict with computed values for confirmation
+    """
+    if not spec_data:
+        return {}
+
+    hierarchy = spec_data.get("hierarchy", {})
+    root = hierarchy.get("spec-root", {})
+
+    # Calculate progress percentage
+    total = root.get("total_tasks", 0)
+    completed = root.get("completed_tasks", 0)
+    progress_pct = int((completed / total * 100)) if total > 0 else 0
+
+    # Determine current phase (first in_progress, or first pending if none)
+    current_phase = None
+    for key, node in hierarchy.items():
+        if node.get("type") == "phase":
+            if node.get("status") == "in_progress":
+                current_phase = key
+                break
+            elif current_phase is None and node.get("status") == "pending":
+                current_phase = key
+
+    # Determine overall status based on progress
+    if total == 0:
+        status = "pending"
+    elif completed == total:
+        status = "completed"
+    elif completed > 0:
+        status = "in_progress"
+    else:
+        status = "pending"
+
+    # Check if any task is blocked - if so, spec is blocked
+    for node in hierarchy.values():
+        if node.get("status") == "blocked":
+            status = "blocked"
+            break
+
+    # Update top-level fields (canonical location)
+    spec_data["progress_percentage"] = progress_pct
+    spec_data["current_phase"] = current_phase
+    spec_data["status"] = status
+
+    # Update last_updated timestamp
+    spec_data["last_updated"] = datetime.now(timezone.utc).isoformat().replace("+00:00", "Z")
+
+    return {
+        "progress_percentage": progress_pct,
+        "current_phase": current_phase,
+        "status": status
+    }

foundry_mcp/core/prometheus.py

@@ -159,7 +159,6 @@ class PrometheusExporter:
         # Manifest/discovery metrics
         self._manifest_tokens: Any = None
         self._manifest_tool_count: Any = None
-        self._feature_flag_state: Any = None
 
         # Health check metrics
         self._health_status: Any = None
@@ -236,11 +235,6 @@ class PrometheusExporter:
                 "Tool count for the advertised tool manifest",
                 ["manifest"],  # unified|legacy
             )
-            self._feature_flag_state = Gauge(
-                f"{ns}_feature_flag_state",
-                "Feature flag state (1=enabled, 0=disabled)",
-                ["flag"],
-            )
 
             # Health check metrics
             self._health_status = Gauge(
@@ -396,13 +390,6 @@ class PrometheusExporter:
         self._manifest_tokens.labels(manifest=manifest_label).set(int(tokens))
         self._manifest_tool_count.labels(manifest=manifest_label).set(int(tool_count))
 
-    def record_feature_flag_state(self, flag: str, enabled: bool) -> None:
-        """Record feature flag enabled/disabled state."""
-        if not self.is_enabled():
-            return
-
-        self._feature_flag_state.labels(flag=flag).set(1 if enabled else 0)
-
     # -------------------------------------------------------------------------
     # Health Check Metrics
     # -------------------------------------------------------------------------

foundry_mcp/core/prompts/fidelity_review.py

@@ -9,6 +9,7 @@ Prompt IDs (PromptTemplate-based):
     - FIDELITY_REVIEW_V1: Main 6-section fidelity review prompt
     - FIDELITY_DEVIATION_ANALYSIS_V1: Analyze identified deviations
     - FIDELITY_COMPLIANCE_SUMMARY_V1: Generate compliance summary
+    - FIDELITY_SYNTHESIS_PROMPT_V1: Multi-model response synthesis
 
 Legacy Prompt IDs (string templates for backward compatibility):
     - review_task: Compare task implementation against spec requirements
@@ -66,6 +67,65 @@ FIDELITY_RESPONSE_SCHEMA = """{
 }"""
 
 
+# JSON response schema for synthesized multi-model fidelity reviews
+FIDELITY_SYNTHESIZED_RESPONSE_SCHEMA = """{
+  "verdict": "pass|fail|partial|unknown",
+  "verdict_consensus": {
+    "votes": {
+      "pass": ["model names that voted pass"],
+      "fail": ["model names that voted fail"],
+      "partial": ["model names that voted partial"],
+      "unknown": ["model names that voted unknown"]
+    },
+    "agreement_level": "strong|moderate|weak|conflicted",
+    "notes": "Explanation of verdict determination"
+  },
+  "summary": "Synthesized overall findings.",
+  "requirement_alignment": {
+    "answer": "yes|no|partial",
+    "details": "Synthesized alignment assessment.",
+    "model_agreement": "unanimous|majority|split"
+  },
+  "success_criteria": {
+    "met": "yes|no|partial",
+    "details": "Synthesized verification status.",
+    "model_agreement": "unanimous|majority|split"
+  },
+  "deviations": [
+    {
+      "description": "Merged deviation description",
+      "justification": "Combined rationale",
+      "severity": "critical|high|medium|low",
+      "identified_by": ["model names that identified this"],
+      "agreement": "unanimous|majority|single"
+    }
+  ],
+  "test_coverage": {
+    "status": "sufficient|insufficient|not_applicable",
+    "details": "Synthesized test assessment",
+    "model_agreement": "unanimous|majority|split"
+  },
+  "code_quality": {
+    "issues": ["Merged quality concerns with model attribution"],
+    "details": "Synthesized commentary"
+  },
+  "documentation": {
+    "status": "adequate|inadequate|not_applicable",
+    "details": "Synthesized doc assessment",
+    "model_agreement": "unanimous|majority|split"
+  },
+  "issues": ["Deduplicated issues with model attribution"],
+  "recommendations": ["Prioritized actionable steps"],
+  "synthesis_metadata": {
+    "models_consulted": ["all model names"],
+    "models_succeeded": ["successful model names"],
+    "models_failed": ["failed model names"],
+    "synthesis_provider": "model that performed synthesis",
+    "agreement_level": "strong|moderate|weak|conflicted"
+  }
+}"""
+
+
 # =============================================================================
 # Severity Categorization Keywords
 # =============================================================================
@@ -166,6 +226,9 @@ CRITICAL CONSTRAINTS:
 - This is a READ-ONLY review - you MUST NOT write, create, or modify ANY files
 - Execute code or commands - ANALYSIS ONLY
 - Provide findings as structured JSON in your response
+- Do NOT focus on ownership, responsibility, or team assignment concerns
+- Avoid feedback like "who owns", "who verifies", "who is responsible for"
+- Focus on technical requirements and verification steps themselves, not who performs them
 
 Focus on:
 1. Requirement alignment - Does implementation match spec?
@@ -430,6 +493,82 @@ Respond with valid JSON:
 )
 
 
+# Multi-model synthesis prompt - consolidates multiple fidelity reviews
+FIDELITY_SYNTHESIS_PROMPT_V1 = PromptTemplate(
+    id="FIDELITY_SYNTHESIS_PROMPT_V1",
+    version="1.0",
+    system_prompt="""You are an expert at synthesizing multiple fidelity review results.
+Your task is to consolidate diverse perspectives into actionable consensus while preserving JSON format.
+
+Guidelines:
+- Attribute findings to specific models using the identified_by field
+- Merge similar deviations, noting which models identified each
+- Resolve verdict disagreements using majority vote or escalate to "partial" on conflict
+- Preserve unique insights from each model
+- Output valid JSON matching the required schema exactly
+- Do NOT focus on ownership, responsibility, or team assignment concerns
+- Focus on technical requirements and verification steps themselves, not who performs them""",
+    user_template="""You are synthesizing {num_models} independent AI fidelity reviews.
+
+**Specification:** {spec_title} (`{spec_id}`)
+**Review Scope:** {review_scope}
+
+**Your Task:** Read all JSON reviews below and create a unified synthesis.
+
+## Individual Model Reviews
+
+{model_reviews}
+
+## Synthesis Requirements
+
+1. **Verdict Consensus:**
+   - Count votes for each verdict (pass/fail/partial/unknown)
+   - Use majority vote for final verdict
+   - If tied or conflicted, use "partial" and note disagreement
+   - Record agreement_level: "strong" (all agree), "moderate" (majority agrees), "weak" (slight majority), "conflicted" (tied/split)
+
+2. **Deviation Merging:**
+   - Group similar deviations across models by description
+   - Use highest severity when models disagree on severity
+   - Track which models identified each deviation in identified_by array
+   - Mark agreement: "unanimous" (all models), "majority" (>50%), "single" (one model)
+
+3. **Issue Consolidation:**
+   - Deduplicate issues across models
+   - Preserve unique insights
+   - Note model agreement level for each finding
+
+4. **Attribution Rules:**
+   - "unanimous" = all successful models agree
+   - "majority" = >50% of successful models agree
+   - "single" = only one model identified this
+
+### Required Response Format
+
+Respond **only** with valid JSON matching the schema below. Do not include Markdown, prose, or additional commentary outside the JSON object.
+
+```json
+{response_schema}
+```
+
+Rules:
+- Use lowercase values for enumerated fields (verdict, status, severity, etc.)
+- Keep arrays as arrays (use [] when empty)
+- Populate identified_by with actual model names from the reviews
+- Never omit required fields from the schema
+- Use the actual provider names from the reviews (e.g., "gemini", "codex", "claude")""",
+    required_context=["spec_id", "spec_title", "review_scope", "num_models", "model_reviews"],
+    optional_context=["response_schema"],
+    metadata={
+        "workflow": "fidelity_review",
+        "author": "system",
+        "category": "synthesis",
+        "output_format": "json",
+        "description": "Multi-model fidelity review synthesis",
+    },
+)
+
+
 # =============================================================================
 # Template Registry (PromptTemplate-based)
 # =============================================================================
@@ -439,6 +578,7 @@ FIDELITY_REVIEW_TEMPLATES: Dict[str, PromptTemplate] = {
     "FIDELITY_REVIEW_V1": FIDELITY_REVIEW_V1,
     "FIDELITY_DEVIATION_ANALYSIS_V1": FIDELITY_DEVIATION_ANALYSIS_V1,
     "FIDELITY_COMPLIANCE_SUMMARY_V1": FIDELITY_COMPLIANCE_SUMMARY_V1,
+    "FIDELITY_SYNTHESIS_PROMPT_V1": FIDELITY_SYNTHESIS_PROMPT_V1,
 }
 
 
@@ -475,7 +615,7 @@ class FidelityReviewPromptBuilder(PromptBuilder):
         Args:
             prompt_id: Template identifier. Supports:
                 - PromptTemplate IDs: FIDELITY_REVIEW_V1, FIDELITY_DEVIATION_ANALYSIS_V1,
-                  FIDELITY_COMPLIANCE_SUMMARY_V1
+                  FIDELITY_COMPLIANCE_SUMMARY_V1, FIDELITY_SYNTHESIS_PROMPT_V1
             context: Template context variables
 
         Returns:
@@ -491,9 +631,12 @@ class FidelityReviewPromptBuilder(PromptBuilder):
             # Provide defaults for optional context
             render_context = dict(context)
 
-            # Add response schema default
+            # Add response schema default - use synthesized schema for synthesis prompt
             if "response_schema" not in render_context:
-                render_context["response_schema"] = FIDELITY_RESPONSE_SCHEMA
+                if prompt_id == "FIDELITY_SYNTHESIS_PROMPT_V1":
+                    render_context["response_schema"] = FIDELITY_SYNTHESIZED_RESPONSE_SCHEMA
+                else:
+                    render_context["response_schema"] = FIDELITY_RESPONSE_SCHEMA
 
             # Add empty defaults for optional fields
             if "spec_description" not in render_context:
@@ -531,10 +674,12 @@ __all__ = [
     "FIDELITY_REVIEW_V1",
     "FIDELITY_DEVIATION_ANALYSIS_V1",
     "FIDELITY_COMPLIANCE_SUMMARY_V1",
+    "FIDELITY_SYNTHESIS_PROMPT_V1",
     # Template registries
     "FIDELITY_REVIEW_TEMPLATES",
-    # Response schema
+    # Response schemas
     "FIDELITY_RESPONSE_SCHEMA",
+    "FIDELITY_SYNTHESIZED_RESPONSE_SCHEMA",
     # Severity keywords
     "SEVERITY_KEYWORDS",
     "CRITICAL_KEYWORDS",

foundry_mcp/core/prompts/markdown_plan_review.py

@@ -71,6 +71,7 @@ What the plan does well.
 - Include all sections even if empty (write "None identified" for empty sections)
 - Be specific and actionable in all feedback
 - For clarity issues, use Questions section rather than creating a separate category
+- Do NOT generate feedback about ownership, responsibility, or team assignments (e.g., "who verifies", "who owns", "who is responsible")
 """
 
 
@@ -89,7 +90,10 @@ Guidelines:
 - Propose alternatives when better approaches exist
 - Focus on impact and prioritize feedback by potential consequences
 - Be collaborative, not adversarial
-- Remember: this is an early-stage plan, not a final spec"""
+- Remember: this is an early-stage plan, not a final spec
+- Do NOT focus on ownership, responsibility, or team assignment concerns
+- Avoid feedback like "who owns", "who verifies", "who is responsible for"
+- Focus on technical requirements and verification steps themselves, not who performs them"""
 
 
 # =============================================================================

foundry_mcp/core/prompts/plan_review.py

@@ -78,6 +78,7 @@ What the spec does well.
 - Be specific and actionable in all feedback
 - For clarity issues, use Questions section rather than creating a separate category
 - Attribution: In multi-model reviews, prefix items with "Flagged by [model-name]:" when applicable
+- Do NOT generate feedback about ownership, responsibility, or team assignments (e.g., "who verifies", "who owns", "who is responsible")
 """
 
 
@@ -94,7 +95,10 @@ Guidelines:
 - Ask clarifying questions for ambiguities
 - Propose alternatives when better approaches exist
 - Focus on impact and prioritize feedback by potential consequences
-- Be collaborative, not adversarial"""
+- Be collaborative, not adversarial
+- Do NOT focus on ownership, responsibility, or team assignment concerns
+- Avoid feedback like "who owns", "who verifies", "who is responsible for"
+- Focus on technical requirements and verification steps themselves, not who performs them"""
 
 
 # =============================================================================

foundry_mcp/core/providers/__init__.py

@@ -56,6 +56,7 @@ from foundry_mcp.core.providers.base import (
     ProviderUnavailableError,
     ProviderExecutionError,
     ProviderTimeoutError,
+    ContextWindowError,
     # ABC
     ProviderContext,
 )
@@ -124,6 +125,11 @@ from foundry_mcp.core.providers.validation import (
     reset_rate_limiters,
     # Execution wrapper
     with_validation_and_resilience,
+    # Context window detection
+    CONTEXT_WINDOW_ERROR_PATTERNS,
+    is_context_window_error,
+    extract_token_counts,
+    create_context_window_guidance,
 )
 
 # ---------------------------------------------------------------------------
@@ -160,6 +166,7 @@ __all__ = [
     "ProviderUnavailableError",
     "ProviderExecutionError",
     "ProviderTimeoutError",
+    "ContextWindowError",
     # ABC
     "ProviderContext",
     # === Detection (detectors.py) ===
@@ -222,4 +229,9 @@ __all__ = [
     "reset_rate_limiters",
     # Execution wrapper
     "with_validation_and_resilience",
+    # Context window detection
+    "CONTEXT_WINDOW_ERROR_PATTERNS",
+    "is_context_window_error",
+    "extract_token_counts",
+    "create_context_window_guidance",
 ]

foundry_mcp/core/providers/base.py

@@ -277,6 +277,44 @@ class ProviderTimeoutError(ProviderError):
     """Raised when a provider exceeds its allotted execution time."""
 
 
+class ContextWindowError(ProviderExecutionError):
+    """Raised when prompt exceeds the model's context window limit.
+
+    This error indicates the prompt/context size exceeded what the model
+    can process. It includes token counts to help with debugging and
+    provides actionable guidance for resolution.
+
+    Attributes:
+        prompt_tokens: Estimated tokens in the prompt (if known)
+        max_tokens: Maximum context window size (if known)
+        provider: Provider that raised the error
+        truncation_needed: How many tokens need to be removed
+    """
+
+    def __init__(
+        self,
+        message: str,
+        *,
+        provider: Optional[str] = None,
+        prompt_tokens: Optional[int] = None,
+        max_tokens: Optional[int] = None,
+    ):
+        """Initialize context window error.
+
+        Args:
+            message: Error message describing the issue
+            provider: Provider ID that raised the error
+            prompt_tokens: Number of tokens in the prompt (if known)
+            max_tokens: Maximum tokens allowed (if known)
+        """
+        super().__init__(message, provider=provider)
+        self.prompt_tokens = prompt_tokens
+        self.max_tokens = max_tokens
+        self.truncation_needed = (
+            (prompt_tokens - max_tokens) if prompt_tokens and max_tokens else None
+        )
+
+
 # =============================================================================
 # Lifecycle Hooks
 # =============================================================================
@@ -471,6 +509,7 @@ __all__ = [
     "ProviderUnavailableError",
     "ProviderExecutionError",
     "ProviderTimeoutError",
+    "ContextWindowError",
     # ABC
     "ProviderContext",
 ]