APPENG-3801-B - Agent performance fixes - all agent stages

kustomize/base/exploit-iq-config.yml

@@ -54,45 +54,45 @@ functions:
   cve_checklist:
     _type: cve_checklist
     llm_name: checklist_llm
-  Transitive code search tool:
+  Call Chain Analyzer:
     _type: transitive_code_search
     enable_transitive_search: true
-  Calling Function Name Extractor:
+  Function Caller Finder:
     _type: calling_function_name_extractor
     enable_functions_usage_search: true
-  Package and Function Locator:
+  Function Locator:
     _type: package_and_function_locator
-  Container Image Code QA System:
+  Code Semantic Search:
     _type: local_vdb_retriever
     embedder_name: nim_embedder
     llm_name: code_vdb_retriever_llm
     vdb_type: code
     return_source_documents: false
-  Container Image Developer Guide QA System:
+  Docs Semantic Search:
     _type: local_vdb_retriever
     embedder_name: nim_embedder
     llm_name: doc_vdb_retriever_llm
     vdb_type: doc
     return_source_documents: false
-  Lexical Search Container Image Code QA System:
+  Code Keyword Search:
     _type: lexical_code_search
     top_k: 5
-  Internet Search:
+  CVE Web Search:
      _type: serp_wrapper
      max_retries: 5
-  Container Image Analysis Data:
+  Container Analysis Data:
     _type: container_image_analysis_data
   cve_agent_executor:
     _type: cve_agent_executor
     llm_name: cve_agent_executor_llm
     tool_names:
-      - Container Image Code QA System
-      - Container Image Developer Guide QA System
-      - Lexical Search Container Image Code QA System  # Uncomment to enable lexical search
-      - Internet Search
-      - Transitive code search tool
-      - Calling Function Name Extractor
-      - Package and Function Locator
+      - Code Semantic Search
+      - Docs Semantic Search
+      - Code Keyword Search
+      - CVE Web Search
+      - Call Chain Analyzer
+      - Function Caller Finder
+      - Function Locator
     max_concurrency: null
     max_iterations: 10
     prompt_examples: false
@@ -106,10 +106,10 @@ functions:
     skip: false
     llm_name: generate_cvss_llm
     tool_names:
-      - Container Image Code QA System
-      - Container Image Developer Guide QA System
-      - Lexical Search Container Image Code QA System  # Uncomment to enable lexical search
-      - Container Image Analysis Data
+      - Code Semantic Search
+      - Docs Semantic Search
+      - Code Keyword Search
+      - Container Analysis Data
     max_concurrency: null
     max_iterations: 10
     prompt_examples: true

kustomize/config-http-openai-local.yml

@@ -55,45 +55,45 @@ functions:
   cve_checklist:
     _type: cve_checklist
     llm_name: checklist_llm
-  Transitive code search tool:
+  Call Chain Analyzer:
     _type: transitive_code_search
     enable_transitive_search: true
-  Calling Function Name Extractor:
+  Function Caller Finder:
     _type: calling_function_name_extractor
     enable_functions_usage_search: true
-  Package and Function Locator:
+  Function Locator:
     _type: package_and_function_locator
-  Container Image Code QA System:
+  Code Semantic Search:
     _type: local_vdb_retriever
     embedder_name: nim_embedder
     llm_name: code_vdb_retriever_llm
     vdb_type: code
     return_source_documents: false
-  Container Image Developer Guide QA System:
+  Docs Semantic Search:
     _type: local_vdb_retriever
     embedder_name: nim_embedder
     llm_name: doc_vdb_retriever_llm
     vdb_type: doc
     return_source_documents: false
-  Lexical Search Container Image Code QA System:
+  Code Keyword Search:
     _type: lexical_code_search
     top_k: 5
-  Internet Search:
+  CVE Web Search:
      _type: serp_wrapper
      max_retries: 5
-  Container Image Analysis Data:
+  Container Analysis Data:
     _type: container_image_analysis_data
   cve_agent_executor:
     _type: cve_agent_executor
     llm_name: cve_agent_executor_llm
     tool_names:
-      - Container Image Code QA System
-      - Container Image Developer Guide QA System
-      - Lexical Search Container Image Code QA System  # Uncomment to enable lexical search
-      - Internet Search
-      - Transitive code search tool
-      - Calling Function Name Extractor
-      - Package and Function Locator
+      - Code Semantic Search
+      - Docs Semantic Search
+      - Code Keyword Search
+      - CVE Web Search
+      - Call Chain Analyzer
+      - Function Caller Finder
+      - Function Locator
     max_concurrency: null
     max_iterations: 10
     prompt_examples: false
@@ -107,10 +107,10 @@ functions:
     skip: false
     llm_name: generate_cvss_llm
     tool_names:
-      - Container Image Code QA System
-      - Container Image Developer Guide QA System
-      - Lexical Search Container Image Code QA System  # Uncomment to enable lexical search
-      - Container Image Analysis Data
+      - Code Semantic Search
+      - Docs Semantic Search
+      - Code Keyword Search
+      - Container Analysis Data
     max_concurrency: null
     max_iterations: 10
     prompt_examples: true
@@ -124,10 +124,20 @@ functions:
   cve_justify:
     _type: cve_justify
     llm_name: justify_llm
+  # cve_file_output:
+  #   _type: cve_file_output
+  #   file_path: .tmp/output.json
+  #   markdown_dir: .tmp/vulnerability_markdown_reports
+  #   overwrite: true
   cve_http_output:
     _type: cve_http_output
     url: http://localhost:8080
     endpoint: /reports
+  cve_file_output:
+    _type: cve_file_output
+    file_path: .tmp/output.json
+    markdown_dir: .tmp/vulnerability_markdown_reports
+    overwrite: true
   cve_calculate_intel_score:
     _type: cve_calculate_intel_score
     llm_name: intel_source_score_llm
@@ -138,55 +148,55 @@ functions:
 llms:
   checklist_llm:
     _type: openai
-    api_key: "EMPTY"
+    api_key: ${OPENAI_API_KEY:-EMPTY}
     base_url: ${NVIDIA_API_BASE:-https://integrate.api.nvidia.com/v1}
     model_name: ${CHECKLIST_MODEL_NAME:-meta/llama-3.1-70b-instruct}
     temperature: 0.0
     max_tokens: 2000
     top_p: 0.01
   code_vdb_retriever_llm:
     _type: openai
-    api_key: "EMPTY"
+    api_key: ${OPENAI_API_KEY:-EMPTY}
     base_url: ${NVIDIA_API_BASE:-https://integrate.api.nvidia.com/v1}
     model_name: ${CODE_VDB_RETRIEVER_MODEL_NAME:-meta/llama-3.1-70b-instruct}
     temperature: 0.0
     max_tokens: 2000
     top_p: 0.01
   doc_vdb_retriever_llm:
     _type: openai
-    api_key: "EMPTY"
+    api_key: ${OPENAI_API_KEY:-EMPTY}
     base_url: ${NVIDIA_API_BASE:-https://integrate.api.nvidia.com/v1}
     model_name: ${DOC_VDB_RETRIEVER_MODEL_NAME:-meta/llama-3.1-70b-instruct}
     temperature: 0.0
     max_tokens: 2000
     top_p: 0.01
   cve_agent_executor_llm:
     _type: openai
-    api_key: "EMPTY"
+    api_key: ${OPENAI_API_KEY:-EMPTY}
     base_url: ${NVIDIA_API_BASE:-https://integrate.api.nvidia.com/v1}
     model_name: ${CVE_AGENT_EXECUTOR_MODEL_NAME:-meta/llama-3.1-70b-instruct}
     temperature: 0.0
     max_tokens: 2000
     top_p: 0.01
   generate_cvss_llm:
     _type: openai
-    api_key: "EMPTY"
+    api_key: ${OPENAI_API_KEY:-EMPTY}
     base_url: ${NVIDIA_API_BASE:-https://integrate.api.nvidia.com/v1}
     model_name: ${GENERATE_CVSS_MODEL_NAME:-meta/llama-3.1-70b-instruct}
     temperature: 0.0
     max_tokens: 1024
     top_p: 0.01
   summarize_llm:
     _type: openai
-    api_key: "EMPTY"
+    api_key: ${OPENAI_API_KEY:-EMPTY}
     base_url: ${NVIDIA_API_BASE:-https://integrate.api.nvidia.com/v1}
     model_name: ${SUMMARIZE_MODEL_NAME:-meta/llama-3.1-70b-instruct}
     temperature: 0.0
     max_tokens: 1024
     top_p: 0.01
   justify_llm:
     _type: openai
-    api_key: "EMPTY"
+    api_key: ${OPENAI_API_KEY:-EMPTY}
     base_url: ${NVIDIA_API_BASE:-https://integrate.api.nvidia.com/v1}
     model_name: ${JUSTIFY_MODEL_NAME:-meta/llama-3.1-70b-instruct}
     temperature: 0.0
@@ -195,7 +205,7 @@ llms:
 
   intel_source_score_llm:
     _type: openai
-    api_key: "EMPTY"
+    api_key: ${OPENAI_API_KEY:-EMPTY}
     base_url: ${NVIDIA_API_BASE:-https://integrate.api.nvidia.com/v1}
     model_name: ${JUSTIFY_MODEL_NAME:-meta/llama-3.1-70b-instruct}
     temperature: 0.0
@@ -222,7 +232,7 @@ workflow:
   cve_generate_cvss_name: cve_generate_cvss
   cve_summarize_name: cve_summarize
   cve_justify_name: cve_justify
-  cve_output_config_name: cve_http_output
+  cve_output_config_name: cve_file_output
 
 eval:
   general: