🐛 Bugfix: Enhance prompt generation with knowledge base display names part2 by Zhi-a · Pull Request #2813 · ModelEngine-Group/nexent

Zhi-a · 2026-04-17T09:23:50Z

✨ Enhance prompt generation with knowledge base display names
智能体提示词，使用工具选择的知识库名称，而不是编造

Added knowledge_base_display_names to the GeneratePromptRequest model to allow frontend-configured names for knowledge bases.
Updated backend functions to utilize these display names, improving few-shot example generation without requiring database lookups.
Modified frontend components to capture and pass knowledge base display names during prompt generation.
Enhanced tests to cover the new functionality and ensure proper integration of knowledge base display names in the prompt generation process.

修改前

修改后，

- Introduced `get_knowledge_name_map_by_index_names` function to retrieve a mapping of index names to their corresponding display names. - Updated `create_agent_config` and `create_tool_config_list` to utilize the new mapping for generating user-friendly summaries. - Enhanced `KnowledgeBaseSearchTool` to support conversion from display names to index names during queries. - Added tests to verify the functionality of the new mapping and its integration within the tool configuration process.

- Added `knowledge_base_display_names` to the `GeneratePromptRequest` model to allow frontend-configured names for knowledge bases. - Updated backend functions to utilize these display names, improving few-shot example generation without requiring database lookups. - Modified frontend components to capture and pass knowledge base display names during prompt generation. - Enhanced tests to cover the new functionality and ensure proper integration of knowledge base display names in the prompt generation process.

Copilot

Pull request overview

This PR adds end-to-end support for using knowledge base display names (frontend-configured) during system prompt generation and tool execution, reducing reliance on database lookups and improving few-shot example correctness.

Changes:

Extends prompt generation APIs/services to accept knowledge_base_display_names (frontend takes precedence over DB-derived names) and inject them into prompt templates.
Adds get_knowledge_name_map_by_index_names() to map internal index_name -> knowledge_name, and uses it to build display_name_to_index_map for KB tool parameter conversion.
Updates frontend tool config + prompt generation flow to collect and pass KB display names; expands tests across backend and SDK.

Reviewed changes

Copilot reviewed 19 out of 19 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
`sdk/nexent/core/tools/knowledge_base_search_tool.py`	Adds `display_name_to_index_map` and converts display names to index names before querying.
`sdk/nexent/core/agents/nexent_agent.py`	Passes `display_name_to_index_map` via metadata after tool instantiation (excluded params).
`backend/database/knowledge_db.py`	Adds `get_knowledge_name_map_by_index_names()` for index-name → display-name lookup.
`backend/services/tool_configuration_service.py`	Builds `display_name_to_index_map` during local tool validation for KB tool.
`backend/services/prompt_service.py`	Accepts `knowledge_base_display_names`, injects into template context, adds DB fallback helper.
`backend/prompts/utils/prompt_generate_zh.yaml`	Adds KB configuration note block for few-shot generation.
`backend/prompts/utils/prompt_generate_en.yaml`	Adds KB configuration note block for few-shot generation.
`backend/consts/model.py`	Extends `GeneratePromptRequest` with `knowledge_base_display_names`.
`backend/apps/prompt_app.py`	Wires request field through to prompt generation streaming endpoint.
`backend/agents/create_agent_info.py`	Uses display names in KB summary and builds `display_name_to_index_map` in tool metadata.
`frontend/types/agentConfig.ts`	Adds `display_names` on tools and `knowledge_base_display_names` in prompt params.
`frontend/app/[locale]/agents/components/agentInfo/AgentGenerateDetail.tsx`	Collects KB display names from tools and sends with prompt generation request.
`frontend/app/[locale]/agents/components/agentConfig/tool/ToolConfigModal.tsx`	Persists selected KB display names on the configured tool in local state.
`test/sdk/core/tools/test_knowledge_base_search_tool.py`	Updates tool instantiation and adds extensive unit tests for new behaviors.
`test/sdk/core/agents/test_nexent_agent.py`	Tests that metadata-driven tool creation sets `display_name_to_index_map`.
`test/backend/services/test_tool_configuration_service.py`	Updates validation tests to account for KB display mapping.
`test/backend/services/test_prompt_service.py`	Adds tests for KB display name injection and DB fallback helper.
`test/backend/database/test_knowledge_db.py`	Adds tests for `get_knowledge_name_map_by_index_names()`.
`test/backend/agents/test_create_agent_info.py`	Updates tool metadata assertions and adds display-name map tests.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-17T09:28:48Z

+    if (Array.isArray(editedAgent.tools)) {
+      for (const tool of editedAgent.tools) {
+        if (typeof tool === "object" && tool.display_names && Array.isArray(tool.display_names)) {
+          knowledgeBaseDisplayNames.push(...tool.display_names);


knowledgeBaseDisplayNames.push(...tool.display_names) can accumulate duplicates (e.g., multiple KB tools or repeated selections), which unnecessarily bloats the payload and prompt. Consider deduplicating while preserving order (e.g., with a Set) before sending knowledge_base_display_names.

Suggested change

if (Array.isArray(editedAgent.tools)) {

for (const tool of editedAgent.tools) {

if (typeof tool === "object" && tool.display_names && Array.isArray(tool.display_names)) {

knowledgeBaseDisplayNames.push(...tool.display_names);

const seenKnowledgeBaseDisplayNames = new Set<string>();

if (Array.isArray(editedAgent.tools)) {

for (const tool of editedAgent.tools) {

if (

typeof tool === "object" &&

tool.display_names &&

Array.isArray(tool.display_names)

) {

for (const displayName of tool.display_names) {

if (!seenKnowledgeBaseDisplayNames.has(displayName)) {

seenKnowledgeBaseDisplayNames.add(displayName);

knowledgeBaseDisplayNames.push(displayName);

}

}

Copilot · 2026-04-17T09:28:48Z

+        """Test that cite_index in results starts from record_ops + index + 1."""
+        mock_results = create_mock_search_result(2)
+        knowledge_base_search_tool.vdb_core.hybrid_search.return_value = mock_results
+
+        # record_ops starts at 1, so cite_index should be 1+0+1=1, 1+1+1=2


The comment explaining the expected cite_index math is inconsistent: 1+0+1 equals 2, but the assertions expect the first cite_index to be 1. Update the comment to match the actual intended formula (cite_index = record_ops + index per implementation) to avoid confusing future readers.

Suggested change

"""Test that cite_index in results starts from record_ops + index + 1."""

mock_results = create_mock_search_result(2)

knowledge_base_search_tool.vdb_core.hybrid_search.return_value = mock_results

# record_ops starts at 1, so cite_index should be 1+0+1=1, 1+1+1=2

"""Test that cite_index in results starts from record_ops + index."""

mock_results = create_mock_search_result(2)

knowledge_base_search_tool.vdb_core.hybrid_search.return_value = mock_results

# record_ops starts at 1, so cite_index should be record_ops + index: 1+0=1, 1+1=2

Copilot · 2026-04-17T09:28:49Z

+        display_name_to_index_map: dict = Field(
+            description="Mapping from display_name (knowledge_name) to index_name",
+            default_factory=dict, exclude=True),
    ):
        """Initialize the KBSearchTool.

        Args:
            top_k (int, optional): Number of results to return. Defaults to 3.
            observer (MessageObserver, optional): Message observer instance. Defaults to None.
+            display_name_to_index_map (dict, optional): Mapping from display_name to index_name.
+                When LLM passes display_name as index_names parameter, it will be converted
+                to the actual index_name for ES queries.

        Raises:
            ValueError: If language is not supported
        """
        super().__init__()
        self.top_k = top_k
        self.observer = observer
        self.vdb_core = vdb_core
        self.index_names = [] if index_names is None else index_names
        self.search_mode = search_mode
        self.embedding_model = embedding_model
        self.rerank = rerank
        self.rerank_model_name = rerank_model_name
        self.rerank_model = rerank_model
+        self.display_name_to_index_map = display_name_to_index_map

        self.record_ops = 1  # To record serial number
        self.running_prompt_zh = "知识库检索中..."
        self.running_prompt_en = "Searching the knowledge base..."


+    def _convert_to_index_names(self, names: List[str]) -> List[str]:
+        """Convert display names (knowledge_name) to index names if necessary.
+
+        When LLM passes display_name as the index_names parameter,
+        this method converts it to the actual index_name for ES queries.
+
+        Args:
+            names: List of names that could be either display_name or index_name
+
+        Returns:
+            List of actual index_names for ES queries
+        """
+        # Handle FieldInfo case (smolagents doesn't expand Field defaults)
+        display_map = self.display_name_to_index_map
+        if isinstance(display_map, FieldInfo):
+            display_map = display_map.default
+        if not display_map:
+            return names


display_name_to_index_map uses Field(default_factory=dict) in a plain Python __init__, so when the caller doesn't pass it, the value is a FieldInfo whose .default is typically PydanticUndefined (because the default comes from default_factory). _convert_to_index_names() then sets display_map = display_map.default, which can lead to a non-dict sentinel and cause failures like TypeError: argument of type 'PydanticUndefinedType' is not iterable when doing name in display_map. Consider normalizing in __init__ (e.g., if it's FieldInfo, use {} / call default_factory), and/or make the signature default None and set {} explicitly to avoid storing FieldInfo on the instance.

Copilot · 2026-04-17T09:28:49Z

+            # Build display_name to index_name mapping for LLM parameter conversion
+            index_names = instantiation_params.get("index_names", [])
+            display_name_to_index_map = {}
+            if index_names:
+                knowledge_name_map = get_knowledge_name_map_by_index_names(index_names)
+                for idx_name, kb_name in knowledge_name_map.items():
+                    display_name_to_index_map[kb_name] = idx_name


index_names = instantiation_params.get("index_names", []) can be a JSON string (the codebase already handles this case elsewhere, e.g. in prompt_service.get_knowledge_base_display_names). Passing a string into get_knowledge_name_map_by_index_names() will produce incorrect queries (.in_(index_names) iterates characters) or runtime errors. Normalize index_names to List[str] here (parse JSON string, and/or wrap a single string into a list) before building display_name_to_index_map.

Copilot · 2026-04-17T09:28:49Z

+    # Get knowledge base display names for few-shot examples
+    # Priority: frontend-provided > database query
+    if knowledge_base_display_names:
+        logger.debug(f"Using frontend-provided knowledge base display names: {knowledge_base_display_names}")
+    else:
+        knowledge_base_display_names = get_knowledge_base_display_names(
+            tool_info_list=tool_info_list,
+            agent_id=agent_id,
+            tenant_id=tenant_id
+        )
+        logger.debug(f"Using database query for knowledge base display names: {knowledge_base_display_names}")


The check if knowledge_base_display_names: treats an empty list the same as None, so a client that intentionally passes [] (to explicitly override DB values) will unexpectedly fall back to the database lookup. If the intent is “frontend-provided takes precedence even if empty”, use an explicit is not None check (and optionally validate/dedupe the list).

Copilot · 2026-04-17T09:28:50Z

+
+    # Add knowledge base display names for few-shot examples if available
+    if knowledge_base_display_names:
+        kb_names_str = ", ".join(f'"{name}"' for name in knowledge_base_display_names)


kb_names_str = ", ".join(f'"{name}"' for name in knowledge_base_display_names) will generate malformed examples if a display name contains quotes, backslashes, or newlines. Since these values can come from the frontend/database, consider producing a JSON-escaped representation (e.g., via json.dumps) or otherwise escaping/quoting robustly before injecting into the prompt template.

Suggested change

kb_names_str = ", ".join(f'"{name}"' for name in knowledge_base_display_names)

kb_names_str = ", ".join(json.dumps(name, ensure_ascii=False) for name in knowledge_base_display_names)

Copilot · 2026-04-17T09:28:50Z

+def get_knowledge_name_map_by_index_names(index_names: List[str]) -> Dict[str, str]:
+    """
+    Get a mapping from index_name to knowledge_name (display name) for the given index_names.
+    Used to build user-friendly knowledge base summaries in prompts.
+
+    Args:
+        index_names: List of internal index names
+
+    Returns:
+        Dict[str, str]: Mapping of index_name -> knowledge_name.
+                       If a knowledge base is not found in the database,
+                       the index_name itself is used as the fallback value.
+    """
+    if not index_names:
+        return {}
+
+    try:
+        with get_db_session() as session:
+            result = session.query(
+                KnowledgeRecord.index_name,
+                KnowledgeRecord.knowledge_name
+            ).filter(
+                KnowledgeRecord.index_name.in_(index_names),
+                KnowledgeRecord.delete_flag != 'Y'
+            ).all()


get_knowledge_name_map_by_index_names() queries by index_name only and does not filter by tenant_id. In a multi-tenant system, this can leak knowledge base names across tenants if index_name is not guaranteed globally unique (or if callers can supply arbitrary index_names). Consider adding a tenant_id parameter and filtering on it in the query, then update callers to pass the tenant context.

…into develop_fix_kb_4 # Conflicts: # backend/prompts/utils/prompt_generate_zh.yaml # test/backend/agents/test_create_agent_info.py

…into develop_fix_kb_3 # Conflicts: # test/backend/agents/test_create_agent_info.py

…into develop_fix_kb_4 # Conflicts: # frontend/app/[locale]/agents/components/agentInfo/AgentGenerateDetail.tsx

codecov · 2026-04-21T06:46:57Z

Codecov Report

❌ Patch coverage is 87.23404% with 6 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
backend/services/prompt_service.py	86.95%	3 Missing and 3 partials ⚠️

📢 Thoughts on this report? Let us know!

…up/nexent into develop_fix_kb_4 # Conflicts: # test/backend/agents/test_create_agent_info.py

…into develop_fix_kb_4 # Conflicts: # test/backend/agents/test_create_agent_info.py

… part2

Dallas98 · 2026-04-27T08:08:07Z

            tool_ids=prompt_request.tool_ids,
-            sub_agent_ids=prompt_request.sub_agent_ids
+            sub_agent_ids=prompt_request.sub_agent_ids,
+            knowledge_base_display_names=prompt_request.knowledge_base_display_names


这个对齐为啥这么奇怪啊

Zhi-a added 2 commits April 15, 2026 15:01

Copilot AI review requested due to automatic review settings April 17, 2026 09:23

Zhi-a requested review from Dallas98 and WMC001 as code owners April 17, 2026 09:23

Copilot started reviewing on behalf of Zhi-a April 17, 2026 09:24 View session

Copilot AI reviewed Apr 17, 2026

View reviewed changes

Merge branch 'develop' of https://github.com/ModelEngine-Group/nexent …

5cba40e

…into develop_fix_kb_4 # Conflicts: # backend/prompts/utils/prompt_generate_zh.yaml # test/backend/agents/test_create_agent_info.py

Zhi-a changed the title ~~✨ Enhance prompt generation with knowledge base display names~~ ✨ Enhance prompt generation with knowledge base display names part2 Apr 20, 2026

Zhi-a added 3 commits April 21, 2026 14:13

Merge branch 'develop' of https://github.com/ModelEngine-Group/nexent …

bb402b8

…into develop_fix_kb_3 # Conflicts: # test/backend/agents/test_create_agent_info.py

Merge branch 'develop' of https://github.com/ModelEngine-Group/nexent …

edf66b4

…into develop_fix_kb_4 # Conflicts: # frontend/app/[locale]/agents/components/agentInfo/AgentGenerateDetail.tsx

✨ Enhance prompt generation with knowledge base display names part2

9d2929e

Zhi-a added 2 commits April 23, 2026 16:02

✨ Add display name to index name mapping for KnowledgeBaseSearchTool

65f7dec

Merge branch 'develop_fix_kb_3' of https://github.com/ModelEngine-Gro…

399f379

…up/nexent into develop_fix_kb_4 # Conflicts: # test/backend/agents/test_create_agent_info.py

Zhi-a changed the title ~~✨ Enhance prompt generation with knowledge base display names part2~~ 🐛 Bugfix: Enhance prompt generation with knowledge base display names part2 Apr 25, 2026

Zhi-a added 2 commits April 25, 2026 14:42

Merge branch 'develop' of https://github.com/ModelEngine-Group/nexent …

95ceeef

…into develop_fix_kb_4 # Conflicts: # test/backend/agents/test_create_agent_info.py

🐛 Bugfix: Enhance prompt generation with knowledge base display names…

66dd82a

… part2

Dallas98 approved these changes Apr 27, 2026

View reviewed changes

Dallas98 merged commit df61b72 into develop Apr 27, 2026
15 of 16 checks passed

Zhi-a deleted the develop_fix_kb_4 branch April 28, 2026 07:08

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

🐛 Bugfix: Enhance prompt generation with knowledge base display names part2#2813

🐛 Bugfix: Enhance prompt generation with knowledge base display names part2#2813
Dallas98 merged 10 commits intodevelopfrom
develop_fix_kb_4

Zhi-a commented Apr 17, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 17, 2026

Uh oh!

Copilot AI Apr 17, 2026

Uh oh!

Copilot AI Apr 17, 2026

Uh oh!

Copilot AI Apr 17, 2026

Uh oh!

Copilot AI Apr 17, 2026

Uh oh!

Copilot AI Apr 17, 2026

Uh oh!

Copilot AI Apr 17, 2026

Uh oh!

codecov Bot commented Apr 21, 2026 •

edited

Loading

Uh oh!

Dallas98 Apr 27, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

	kb_names_str = ", ".join(f'"{name}"' for name in knowledge_base_display_names)
	kb_names_str = ", ".join(json.dumps(name, ensure_ascii=False) for name in knowledge_base_display_names)

Conversation

Zhi-a commented Apr 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 17, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 17, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 17, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 17, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 17, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 17, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 17, 2026

Choose a reason for hiding this comment

Uh oh!

codecov Bot commented Apr 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Dallas98 Apr 27, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Zhi-a commented Apr 17, 2026 •

edited

Loading

codecov Bot commented Apr 21, 2026 •

edited

Loading