feat(pr-review): implement memory storage and retrieval in PR review workflows

Enhance PR review workflows by adding memory storage and retrieval capabilities. This includes storing findings and initial concerns during different phases and utilizing them in the synthesis phase to generate detailed reports. This change enables more comprehensive and iterative analysis across multiple review iterations.

Author: ibacalu

src/workflows_mcp/templates/agents/pr-review/agent-investigation-loop/investigation-sub-workflow/main.yaml

@@ -338,6 +338,31 @@ blocks:
             description: "Confidence level in this analysis"
         required: [summary, issues, severity, suggestions, needs_further_investigation, confidence]
 
+  # ============================================================================
+  # MEMORY: Store file findings for accumulation across iterations
+  # ============================================================================
+  - id: store_file_findings
+    type: Workflow
+    description: Store file analysis findings in memory for accumulation.
+    condition: "{{ inputs.state | trim != '' }}"
+    depends_on:
+      - analyze_target
+      - extract_target_path
+    inputs:
+      workflow: agent-state-management
+      inputs:
+        state: "{{ inputs.state }}"
+        op: memory
+        memory_op: append
+        memory_key: "findings.by_file"
+        memory_value: |
+          {
+            "path": "{{ (blocks.extract_target_path.outputs.stdout | fromjson).path }}",
+            "analysis": {{ blocks.analyze_target.outputs.response | tojson }},
+            "depth": {{ inputs.investigation_depth }}
+          }
+        caller: "pr-review:investigation-sub-workflow"
+
   # ============================================================================
   # RECURSIVE INVESTIGATION: Process related files if needed
   # ============================================================================
@@ -382,7 +407,10 @@ blocks:
     condition: "{{ inputs.state and inputs.state != '' and blocks.track_file_start.succeeded }}"
     depends_on:
       - analyze_target
-      - investigate_related_files
+      - block: store_file_findings
+        required: false
+      - block: investigate_related_files
+        required: false
     inputs:
       workflow: agent-state-management
       inputs:

src/workflows_mcp/templates/agents/pr-review/agent-investigation-loop/process-investigation-targets/main.yaml

@@ -179,6 +179,31 @@ blocks:
             description: "New targets to investigate if continuation is needed"
         required: [confidence, needs_more_investigation, findings, conclusion]
 
+  # ============================================================================
+  # MEMORY: Store iteration findings for accumulation
+  # ============================================================================
+  - id: store_iteration_findings
+    type: Workflow
+    description: Store this iteration's findings in memory for accumulation.
+    condition: "{{ inputs.state | trim != '' }}"
+    depends_on:
+      - gather_results
+    inputs:
+      workflow: agent-state-management
+      inputs:
+        state: "{{ inputs.state }}"
+        op: memory
+        memory_op: append
+        memory_key: "iterations.history"
+        memory_value: |
+          {
+            "confidence": {{ blocks.gather_results.outputs.response.confidence | default(0) }},
+            "findings": {{ blocks.gather_results.outputs.response.findings | default([]) | tojson }},
+            "conclusion": "{{ blocks.gather_results.outputs.response.conclusion | default('') }}",
+            "needs_more": {{ blocks.gather_results.outputs.response.needs_more_investigation | default(false) | tojson }}
+          }
+        caller: "pr-review:process-targets"
+
   # ============================================================================
   # ITERATION: Advance iteration counter and check if capped
   # ============================================================================
@@ -187,6 +212,8 @@ blocks:
     description: Advance iteration counter and check if safety cap reached.
     depends_on:
       - gather_results
+      - block: store_iteration_findings
+        required: false
     condition: "{{ inputs.state | trim != '' and inputs.task_id | trim != '' }}"
     inputs:
       workflow: agent-state-management

src/workflows_mcp/templates/agents/pr-review/initial-assessment/main.yaml

@@ -85,6 +85,9 @@ blocks:
     description: |
       Consolidate PR analysis, risk assessment, and identification of
       investigation targets into a single structured response.
+    depends_on:
+      - block: track_start
+        required: false
     inputs:
       profile: default
       prompt: |
@@ -215,6 +218,41 @@ blocks:
             description: "Brief summary of the change scope"
         required: [pr_intent, risk_level, risk_rationale, investigation_targets]
 
+  # ============================================================================
+  # MEMORY: Store initial concerns for synthesis phase
+  # ============================================================================
+  - id: store_initial_concerns
+    type: Workflow
+    description: Store initial concerns in memory for synthesis phase access.
+    condition: "{{ inputs.state | trim != '' }}"
+    depends_on:
+      - consolidated_pr_assessment
+    inputs:
+      workflow: agent-state-management
+      inputs:
+        state: "{{ inputs.state }}"
+        op: memory
+        memory_op: set
+        memory_key: "synthesis.initial_concerns"
+        memory_value: "{{ blocks.consolidated_pr_assessment.outputs.response.initial_concerns | default([]) | tojson }}"
+        caller: "pr-review:initial-assessment"
+
+  - id: store_investigation_targets
+    type: Workflow
+    description: Store investigation targets in memory for reference.
+    condition: "{{ inputs.state | trim != '' }}"
+    depends_on:
+      - consolidated_pr_assessment
+    inputs:
+      workflow: agent-state-management
+      inputs:
+        state: "{{ inputs.state }}"
+        op: memory
+        memory_op: set
+        memory_key: "investigation.targets.initial"
+        memory_value: "{{ blocks.consolidated_pr_assessment.outputs.response.investigation_targets | tojson }}"
+        caller: "pr-review:initial-assessment"
+
   # ============================================================================
   # STATE TRACKING: Mark phase complete with captured data
   # ============================================================================
@@ -224,6 +262,10 @@ blocks:
     condition: "{{ inputs.state | trim != '' and blocks.track_start.succeeded }}"
     depends_on:
       - consolidated_pr_assessment
+      - block: store_initial_concerns
+        required: false
+      - block: store_investigation_targets
+        required: false
     inputs:
       workflow: agent-state-management
       inputs:

src/workflows_mcp/templates/agents/pr-review/main.yaml

@@ -90,6 +90,32 @@ blocks:
         ai: true
         audit: true
 
+  # ==========================================================================
+  # MEMORY INITIALIZATION: Clear memory for fresh review
+  # ==========================================================================
+  - id: init_memory
+    type: Workflow
+    description: |
+      Initialize/clear memory for this PR review execution.
+      Creates fresh memory file to avoid cross-review contamination.
+    depends_on:
+      - state_management
+    condition: "{{ blocks.state_management.outputs.state | trim != '' }}"
+    inputs:
+      workflow: agent-state-management
+      inputs:
+        state: "{{ blocks.state_management.outputs.state }}"
+        op: memory
+        memory_op: set
+        memory_key: "_initialized"
+        memory_value: |
+          {
+            "trace_id": "{{ blocks.state_management.outputs.trace_id }}",
+            "started_at": "{{ now() }}",
+            "version": 1
+          }
+        caller: "pr-review:main"
+
   # ==========================================================================
   # PHASE 1: Context Gathering
   # Normalizes input (URL vs local path) and outputs consistent repo_path
@@ -101,6 +127,8 @@ blocks:
       Outputs normalized repo_path regardless of input source.
     depends_on:
       - state_management
+      - block: init_memory
+        required: false
     inputs:
       workflow: context-gathering
       inputs:

src/workflows_mcp/templates/agents/pr-review/synthesis/generate-report/main.yaml

@@ -33,6 +33,14 @@ inputs:
     required: false
     default: general
 
+  memory_summary:
+    type: dict
+    description: |
+      Summary of accumulated memory data for report.
+      Contains metrics from memory accumulation across phases.
+    required: false
+    default: {}
+
 blocks:
   - id: generate_report_llm
     type: LLMCall
@@ -61,6 +69,17 @@ blocks:
         General comprehensive review.
         {% endif %}
 
+        ## Investigation Summary
+        {% if inputs.memory_summary and inputs.memory_summary != {} %}
+        - **Initial Concerns Captured:** {{ inputs.memory_summary.initial_concerns_count | default(0) }}
+        - **Files Analyzed:** {{ inputs.memory_summary.files_analyzed | default(0) }}
+        - **Investigation Iterations:** {{ inputs.memory_summary.iterations_completed | default(0) }}
+        - **Total Findings (before deduplication):** {{ inputs.memory_summary.total_findings_from_memory | default(0) }}
+        - **Final Unique Findings:** {{ inputs.memory_summary.deduplicated_findings | default(0) }}
+        {% else %}
+        _No memory summary available_
+        {% endif %}
+
         ## Findings ({{ inputs.filtered_findings | length }} total)
         ```json
         {{ inputs.filtered_findings | tojson }}
@@ -84,11 +103,15 @@ blocks:
            - Highlight {{ inputs.focus }}-related issues prominently
            {% endif %}
 
-        3. **Recommendations**
+        3. **Investigation Summary** (if memory_summary provided)
+           - Include the metrics table from Investigation Summary above
+           - Note if any findings came from initial concerns vs iterations
+
+        4. **Recommendations**
            - Actionable steps to address findings
            - Priority order
 
-        4. **Conclusion**
+        5. **Conclusion**
            - Overall verdict
            - Whether PR is ready to merge