ralphx 0.3.4__py3-none-any.whl → 0.4.0__py3-none-any.whl

ralphx/core/planning_service.py

@@ -64,25 +64,109 @@ This format is machine-parsed. Non-compliance breaks the system.
 # =============================================================================
 
 PLANNING_BEHAVIOR = '''<behavior>
-You are a product planning assistant helping users design software products.
-
-Your role is to:
-1. Understand what the user wants to build
-2. Ask clarifying questions about requirements, users, and scope
-3. Help them think through the architecture and design
-4. Eventually produce a comprehensive design document
-
-Guidelines:
-- Ask focused, specific questions (2-3 at a time max)
-- Summarize your understanding periodically
-- Be conversational but efficient
-- Focus on: problem space, target users, core features, technical constraints
-- When the user is ready, offer to generate the design document
-
-Do NOT:
-- Write code unless explicitly asked
-- Make assumptions without confirming
-- Overwhelm with too many questions at once
+You are an expert product architect and technical consultant helping users design
+software products. You combine deep technical knowledge with strong product sense.
+
+## Your Core Mission
+
+Transform vague ideas into comprehensive, implementable design documents through
+collaborative discovery. You're not just a Q&A bot—you're a thinking partner who
+challenges assumptions, identifies blind spots, and brings industry expertise.
+
+## Discovery Phases
+
+Work through these phases naturally (you don't need to announce them):
+
+### Phase 1: Problem Space Understanding
+- What problem are we solving? Why does it matter?
+- Who experiences this problem? (specific personas, not generic "users")
+- What do they currently do? What's broken about that?
+- What would success look like for them?
+
+### Phase 2: Solution Requirements
+- Core features (MVP vs nice-to-have)
+- User workflows and key interactions
+- Data the system needs to handle
+- Integration points with other systems
+- Constraints: budget, timeline, team skills, existing infrastructure
+
+### Phase 3: Technical Architecture
+- System components and how they communicate
+- Data model and storage choices
+- API design (if applicable)
+- Authentication and authorization approach
+- Third-party services and dependencies
+
+### Phase 4: Infrastructure & Operations
+- Hosting environment (cloud provider, on-prem, hybrid)
+- Deployment strategy (containers, serverless, VMs)
+- Scaling considerations
+- Monitoring and observability
+- Backup and disaster recovery
+
+### Phase 5: Security & Compliance
+- Data sensitivity and protection requirements
+- Authentication mechanisms
+- Compliance requirements (GDPR, HIPAA, SOC2, etc.)
+- Threat model considerations
+
+## How to Ask Questions
+
+Ask 2-4 focused questions at a time. For each question:
+- Explain WHY you're asking (what decision it informs)
+- Offer concrete options when helpful, not just open-ended questions
+- Share your initial thinking or recommendation if you have one
+
+Good: "For authentication, are you thinking OAuth (Google/GitHub login) for simplicity,
+or do you need custom username/password? OAuth is faster to implement and more secure,
+but custom auth gives you more control over the user experience."
+
+Bad: "How do you want to handle authentication?"
+
+## Using Web Search
+
+When you have web search available, use it strategically:
+- Research industry best practices for the specific domain
+- Look up current pricing/capabilities of services you might recommend
+- Find examples of similar products for inspiration
+- Verify technical approaches are current (technologies evolve fast)
+- Look for common pitfalls in this type of application
+
+IMPORTANT: Always tell the user when you're searching and summarize what you learned.
+This builds trust and shows you're doing real research, not just making things up.
+
+## Progressive Refinement
+
+As you learn more:
+- Periodically summarize your understanding ("Here's what I have so far...")
+- Explicitly call out assumptions you're making
+- Revisit earlier decisions if new information changes things
+- Be willing to push back if something doesn't make sense
+
+## Re-engagement Support
+
+If the user is returning to continue or update an existing design doc:
+- Acknowledge what exists and ask what they want to change
+- Don't re-ask questions that are already answered in the doc
+- Focus on the delta—what's new, changed, or needs refinement
+
+## Offering to Generate
+
+When you have enough information for a solid design document:
+- Summarize the key decisions that have been made
+- List any important questions that remain unanswered
+- Offer to generate the design doc (the user will click a button)
+
+The user can generate the document at ANY time—it doesn't have to be "complete."
+Better to generate something and iterate than to wait forever for perfection.
+
+## What NOT To Do
+
+- Don't write code or implementation details (that's for later steps)
+- Don't make major assumptions without confirming
+- Don't be a yes-person—challenge ideas that seem problematic
+- Don't overwhelm with too many questions
+- Don't be generic—tailor advice to their specific situation
 </behavior>'''
 
 ARTIFACT_BEHAVIOR = '''<behavior>
@@ -153,10 +237,17 @@ class PlanningService:
             "format_rules", "request", "design_doc", "guardrails",
         ]
         for tag in dangerous_tags:
-            # Escape opening tags: <tag> or <tag ...>
+            # Escape opening tags with attributes: <tag ...>
+            sanitized = re.sub(
+                rf'<\s*{tag}\s([^>]*)>',
+                rf'&lt;{tag} \1&gt;',
+                sanitized,
+                flags=re.IGNORECASE,
+            )
+            # Escape opening tags without attributes: <tag>
             sanitized = re.sub(
-                rf'<\s*{tag}(\s|>)',
-                rf'&lt;{tag}\1',
+                rf'<\s*{tag}\s*>',
+                rf'&lt;{tag}&gt;',
                 sanitized,
                 flags=re.IGNORECASE,
             )
@@ -264,6 +355,8 @@ Start your response with <design_doc> immediately.
         self,
         messages: list[dict],
         model: str = "sonnet",
+        tools: Optional[list[str]] = None,
+        timeout: int = 180,
     ) -> AsyncIterator[StreamEvent]:
         """Stream Claude's response to the conversation.
 
@@ -272,6 +365,8 @@ Start your response with <design_doc> immediately.
         Args:
             messages: Conversation history.
             model: Model to use (sonnet, opus, haiku).
+            tools: Optional list of tools to enable (e.g., ['WebSearch', 'WebFetch']).
+            timeout: Timeout in seconds (default 180 for web search).
 
         Yields:
             StreamEvent objects as Claude responds.
@@ -286,8 +381,8 @@ Start your response with <design_doc> immediately.
         async for event in adapter.stream(
             prompt=prompt,
             model=model,
-            tools=None,  # No tools for planning chat
-            timeout=120,
+            tools=tools,
+            timeout=timeout,
         ):
             yield event
 

ralphx/core/preview.py

@@ -337,16 +337,14 @@ class PromptPreviewEngine:
         title = item.get("title") or ""
         metadata = item.get("metadata")
         metadata_json = json.dumps(metadata) if metadata else "{}"
-        namespace = item.get("namespace", "unknown")
+        workflow_id = item.get("workflow_id", "unknown")
 
         # Substitution order: most specific first
         result = template.replace("{{input_item.metadata}}", metadata_json)
         result = result.replace("{{input_item.content}}", content)
         result = result.replace("{{input_item.title}}", title)
         result = result.replace("{{input_item}}", content)
-        result = result.replace("{{namespace}}", namespace)
-        # Backward compatibility
-        result = result.replace("{{source_loop}}", namespace)
+        result = result.replace("{{workflow_id}}", workflow_id)
 
         return result
 
@@ -372,28 +370,17 @@ class PromptPreviewEngine:
             return False
         return bool(self.config.item_types.input.source)
 
-    def _get_source_namespace(self) -> Optional[str]:
-        """Get the source namespace for consumer loops.
-
-        For consumer loops, this is the source loop name used as the namespace.
-        """
-        if not self._is_consumer_loop():
-            return None
-        return self.config.item_types.input.source
-
     def _get_sample_item(self) -> Optional[dict]:
-        """Get a sample item from the source namespace for preview.
+        """Get a sample item for preview (for consumer loops).
 
         Returns:
             Sample item dict or None.
         """
-        namespace = self._get_source_namespace()
-        if not namespace:
+        if not self._is_consumer_loop():
             return None
 
-        # Get first completed item from namespace
+        # Get first completed item
         items, _ = self.db.list_work_items(
-            namespace=namespace,
             status="completed",
             limit=1,
         )
@@ -403,7 +390,6 @@ class PromptPreviewEngine:
 
         # Fallback: get any pending item
         items, _ = self.db.list_work_items(
-            namespace=namespace,
             status="pending",
             limit=1,
         )
@@ -414,8 +400,8 @@ class PromptPreviewEngine:
         # Return a placeholder
         return {
             "id": "sample-001",
-            "content": "[Sample item content - no items in namespace]",
-            "namespace": namespace,
+            "content": "[Sample item content - no items available]",
+            "source": source_name,
         }
 
     def _explain_strategy(self) -> str:
@@ -472,15 +458,13 @@ class PromptPreviewEngine:
                 variables["{{input_item.content}}"] = sample_item.get("content", "")[:50] + "..."
                 variables["{{input_item.title}}"] = sample_item.get("title", "")
                 variables["{{input_item.metadata}}"] = "{...}"
-                variables["{{namespace}}"] = sample_item.get("namespace", "")
-                variables["{{source_loop}}"] = sample_item.get("namespace", "")  # Backward compat
+                variables["{{workflow_id}}"] = sample_item.get("workflow_id", "")
             else:
                 variables["{{input_item}}"] = "[Claimed item content]"
                 variables["{{input_item.content}}"] = "[Claimed item content]"
                 variables["{{input_item.title}}"] = "[Claimed item title]"
                 variables["{{input_item.metadata}}"] = "[Claimed item metadata as JSON]"
-                variables["{{namespace}}"] = self._get_source_namespace() or ""
-                variables["{{source_loop}}"] = self._get_source_namespace() or ""  # Backward compat
+                variables["{{workflow_id}}"] = "[Workflow ID]"
 
         # Design doc variable (if configured)
         if self.config.context and self.config.context.design_doc: