version: "1.0.1" name: internal-knowledge description: Discover and retrieve from knowledge bases the current user can access. when_to_use: Use this skill when an assistant answer needs factual context from workspace knowledge bases. provider_type: builtin provider_id: knowledge runtime_type: tool tools:

• list_accessible_knowledge_bases
• retrieve_knowledge

max_calls_per_turn: 20 timeout_seconds: 30 tool_governance: list_accessible_knowledge_bases: tool_id: knowledge.list_accessible skill_id: internal-knowledge domain: knowledge effect: read asset_type: knowledge_base risk_level: low requires_asset_resolution: false reversible: false bulk_sensitive: false external_side_effect: false permission_scopes:

• knowledge:read

default_approval_policy: auto_by_permission_tier allowed_permission_tiers:

• basic
• advanced
• full

audit_required: true idempotency_required: false retrieve_knowledge: tool_id: knowledge.retrieve skill_id: internal-knowledge domain: knowledge effect: read asset_type: knowledge_base risk_level: low requires_asset_resolution: true reversible: false bulk_sensitive: false external_side_effect: false permission_scopes:

• knowledge:read

default_approval_policy: auto_by_permission_tier allowed_permission_tiers:

• basic
• advanced
• full

audit_required: true idempotency_required: false display: icon: library category: knowledge_retrieval scenarios:

• knowledge_research
• office_collaboration

label: en_US: Internal Knowledge zh_Hans: 内部知识库 description: en_US: Designed for answering from company policies, product material, or workspace documents; finds accessible knowledge bases and retrieves relevant facts with source context. zh_Hans: 适用于根据公司制度、产品资料或工作区文档回答问题，可查找有权限访问的知识库并检索相关事实和来源上下文。 when_to_use: en_US: Use when an assistant answer needs facts or source context from accessible knowledge bases. zh_Hans: 当助手回复需要引用可访问知识库中的事实或来源上下文时使用。 tags: en_US:

• Knowledge
• Retrieval

zh_Hans:

• 知识库
• 检索

Internal Knowledge Skill

Use this skill to answer internal assistant questions with context from knowledge bases the current user can access.

Workflow

1. First call list_accessible_knowledge_bases with a short query derived from the user's topic, business domain, or explicitly named knowledge base.
2. Inspect status, fallback_used, and returned knowledge base names/descriptions before selecting datasets:

• If status is success, select only the most relevant returned dataset_id values.
• If status is fallback, treat the candidates as weak matches. Select a dataset only when its name or description is clearly related; otherwise refine the list query once or ask the user which knowledge base/domain to use.
• If status is no_results, answer from conversation context if possible and say no relevant accessible knowledge base was found.

1. Call retrieve_knowledge with the selected dataset_ids and a concise query. Do not use dataset IDs that were not returned by the list tool.
2. Inspect status, source_summary, context_blocks, and scores before answering:

• If retrieved blocks answer the user, answer from those blocks.
• If the blocks are missing, weak, or off-topic, rewrite the query once using clearer entities, synonyms, or constraints and retry with the same relevant datasets.
• After two retrieval attempts, if the relevant answer is still unclear, ask the user for clarification or say that no relevant accessible knowledge was found.

1. When the user asks for original wording, definitions, synopsis text, policy clauses, exact wording, or "what does it say", quote or closely excerpt the retrieved source text first and cite the source. Summarize only when the user asks for a summary or the original text is too long.
2. When retrieved context is used, cite source names from source_summary or retriever_resources when useful.
3. Never expose internal IDs such as dataset ID, document ID, or segment ID to the user.
4. Never invent a knowledge-base answer when no relevant accessible context was found.

Tool Usage

list_accessible_knowledge_bases accepts:

• query: optional search text for narrowing candidate knowledge bases.
• limit: optional maximum number of candidates. Defaults to 20 and is capped at 100.

retrieve_knowledge accepts:

• query: the user question or refined search query.
• dataset_ids: required selected knowledge base IDs returned by the list tool.
• top_k: optional maximum retrieved chunks. Defaults to 5 and is capped at 20.
• retrieval_mode: optional hybrid, vector, or graph. Omit it for the default hybrid mode; use graph only for relationship or entity questions.