foundry-mcp 0.3.3__py3-none-any.whl → 0.7.0__py3-none-any.whl

foundry_mcp/core/llm_config.py

@@ -860,16 +860,18 @@ 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 +880,17 @@ 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 +906,12 @@ 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 +932,9 @@ 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/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/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/claude.py

@@ -14,10 +14,7 @@ import os
 import subprocess
 from typing import Any, Dict, List, Optional, Protocol, Sequence
 
-logger = logging.getLogger(__name__)
-
 from .base import (
-    ModelDescriptor,
     ProviderCapability,
     ProviderContext,
     ProviderExecutionError,
@@ -34,6 +31,8 @@ from .base import (
 from .detectors import detect_provider_availability
 from .registry import register_provider
 
+logger = logging.getLogger(__name__)
+
 DEFAULT_BINARY = "claude"
 DEFAULT_TIMEOUT_SECONDS = 360
 AVAILABILITY_OVERRIDE_ENV = "CLAUDE_CLI_AVAILABLE_OVERRIDE"
@@ -181,34 +180,11 @@ def _default_runner(
     )
 
 
-CLAUDE_MODELS: List[ModelDescriptor] = [
-    ModelDescriptor(
-        id="sonnet",
-        display_name="Sonnet 4.5",
-        capabilities={
-            ProviderCapability.TEXT,
-            ProviderCapability.STREAMING,
-            ProviderCapability.VISION,
-            ProviderCapability.THINKING,
-        },
-        routing_hints={"tier": "default", "description": "Smartest model for daily use"},
-    ),
-    ModelDescriptor(
-        id="haiku",
-        display_name="Haiku 4.5",
-        capabilities={
-            ProviderCapability.TEXT,
-            ProviderCapability.STREAMING,
-        },
-        routing_hints={"tier": "fast", "description": "Fastest model for simple tasks"},
-    ),
-]
-
 CLAUDE_METADATA = ProviderMetadata(
     provider_id="claude",
     display_name="Anthropic Claude CLI",
-    models=CLAUDE_MODELS,
-    default_model="sonnet",
+    models=[],  # Model validation delegated to CLI
+    default_model="opus",
     capabilities={
         ProviderCapability.TEXT,
         ProviderCapability.STREAMING,
@@ -239,24 +215,7 @@ class ClaudeProvider(ProviderContext):
         self._binary = binary or os.environ.get(CUSTOM_BINARY_ENV, DEFAULT_BINARY)
         self._env = env
         self._timeout = timeout or DEFAULT_TIMEOUT_SECONDS
-        self._model = self._ensure_model(model or metadata.default_model or self._first_model_id())
-
-    def _first_model_id(self) -> str:
-        if not self.metadata.models:
-            raise ProviderUnavailableError(
-                "Claude provider metadata is missing model descriptors.",
-                provider=self.metadata.provider_id,
-            )
-        return self.metadata.models[0].id
-
-    def _ensure_model(self, candidate: str) -> str:
-        available = {descriptor.id for descriptor in self.metadata.models}
-        if candidate not in available:
-            raise ProviderExecutionError(
-                f"Unsupported Claude model '{candidate}'. Available: {', '.join(sorted(available))}",
-                provider=self.metadata.provider_id,
-            )
-        return candidate
+        self._model = model or metadata.default_model or "opus"
 
     def _validate_request(self, request: ProviderRequest) -> None:
         """Validate and normalize request, ignoring unsupported parameters."""
@@ -357,7 +316,7 @@ class ClaudeProvider(ProviderContext):
     def _resolve_model(self, request: ProviderRequest) -> str:
         model_override = request.metadata.get("model") if request.metadata else None
         if model_override:
-            return self._ensure_model(str(model_override))
+            return str(model_override)
         return self._model
 
     def _emit_stream_if_requested(self, content: str, *, stream: bool) -> None:

foundry_mcp/core/providers/codex.py

@@ -15,10 +15,7 @@ import os
 import subprocess
 from typing import Any, Dict, List, Optional, Protocol, Sequence, Tuple
 
-logger = logging.getLogger(__name__)
-
 from .base import (
-    ModelDescriptor,
     ProviderCapability,
     ProviderContext,
     ProviderExecutionError,
@@ -35,6 +32,8 @@ from .base import (
 from .detectors import detect_provider_availability
 from .registry import register_provider
 
+logger = logging.getLogger(__name__)
+
 DEFAULT_BINARY = "codex"
 DEFAULT_TIMEOUT_SECONDS = 360
 AVAILABILITY_OVERRIDE_ENV = "CODEX_CLI_AVAILABLE_OVERRIDE"
@@ -228,44 +227,11 @@ def _default_runner(
     )
 
 
-CODEX_MODELS: List[ModelDescriptor] = [
-    ModelDescriptor(
-        id="gpt-5.1-codex",
-        display_name="GPT-5.1 Codex",
-        capabilities={
-            ProviderCapability.TEXT,
-            ProviderCapability.STREAMING,
-            ProviderCapability.FUNCTION_CALLING,
-        },
-        routing_hints={"tier": "primary", "optimized_for": "codex"},
-    ),
-    ModelDescriptor(
-        id="gpt-5.1-codex-mini",
-        display_name="GPT-5.1 Codex Mini",
-        capabilities={
-            ProviderCapability.TEXT,
-            ProviderCapability.STREAMING,
-            ProviderCapability.FUNCTION_CALLING,
-        },
-        routing_hints={"tier": "fast", "optimized_for": "codex"},
-    ),
-    ModelDescriptor(
-        id="gpt-5.1",
-        display_name="GPT-5.1",
-        capabilities={
-            ProviderCapability.TEXT,
-            ProviderCapability.STREAMING,
-            ProviderCapability.FUNCTION_CALLING,
-        },
-        routing_hints={"tier": "general"},
-    ),
-]
-
 CODEX_METADATA = ProviderMetadata(
     provider_id="codex",
     display_name="OpenAI Codex CLI",
-    models=CODEX_MODELS,
-    default_model="gpt-5.1-codex",
+    models=[],  # Model validation delegated to CLI
+    default_model="gpt-5.2",
     capabilities={ProviderCapability.TEXT, ProviderCapability.STREAMING, ProviderCapability.FUNCTION_CALLING},
     security_flags={"writes_allowed": False, "read_only": True, "sandbox": "read-only"},
     extra={
@@ -301,7 +267,7 @@ class CodexProvider(ProviderContext):
         self._binary = binary or os.environ.get(CUSTOM_BINARY_ENV, DEFAULT_BINARY)
         self._env = self._prepare_subprocess_env(env)
         self._timeout = timeout or DEFAULT_TIMEOUT_SECONDS
-        self._model = self._ensure_model(model or metadata.default_model or self._first_model_id())
+        self._model = model or metadata.default_model or "gpt-5.2"
 
     def _prepare_subprocess_env(self, custom_env: Optional[Dict[str, str]]) -> Dict[str, str]:
         """
@@ -323,23 +289,6 @@ class CodexProvider(ProviderContext):
 
         return subprocess_env
 
-    def _first_model_id(self) -> str:
-        if not self.metadata.models:
-            raise ProviderUnavailableError(
-                "Codex provider metadata is missing model descriptors.",
-                provider=self.metadata.provider_id,
-            )
-        return self.metadata.models[0].id
-
-    def _ensure_model(self, candidate: str) -> str:
-        available = {descriptor.id for descriptor in self.metadata.models}
-        if candidate not in available:
-            raise ProviderExecutionError(
-                f"Unsupported Codex model '{candidate}'. Available: {', '.join(sorted(available))}",
-                provider=self.metadata.provider_id,
-            )
-        return candidate
-
     def _validate_request(self, request: ProviderRequest) -> None:
         """Validate and normalize request, ignoring unsupported parameters."""
         unsupported: List[str] = []
@@ -531,7 +480,7 @@ class CodexProvider(ProviderContext):
 
     def _execute(self, request: ProviderRequest) -> ProviderResult:
         self._validate_request(request)
-        model = self._ensure_model(
+        model = (
             str(request.metadata.get("model")) if request.metadata and "model" in request.metadata else self._model
         )
         prompt = self._build_prompt(request)