菜单
Retrieve from knowledge bases explicitly configured on the current Agent.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Extract structured contract fields from existing contract text or already parsed contract content according to a configured field list. Use when the user asks for 合同字段抽取, 合同信息抽取, 合同关键信息提取, 合同台账, contract field extraction, contract metadata extraction, contract ledger data, or extracting parties, amount, dates, term, payment terms, renewal terms, breach clauses, or other configured contract fields. This skill does not parse uploaded files directly; it works only after contract text is available from another system or parser.
Design structured user clarification flows with the existing request_user_input capability so the assistant asks stable, focused questions with usable quick options instead of plain natural-language follow-up. Use when the user or another skill needs 问答交互, 表单交互, 结构化质询, 弹出输入框, 选项框, 单选, 多选, 补充信息, 确认参数, 文件上传提示, 表单填写, questionnaire, form interaction, structured clarification, user input, quick replies, or parameter confirmation.
Plan precise slide-by-slide PPT structures from a topic, text content, requirements, or uploaded-material summaries, and prepare a file-generator-ready PPTX payload for downstream generation. Use when the user asks for PPT planning, PowerPoint outline, slide planning, page-by-page PPT requirements, presentation storyboard, report deck structure, pitch deck outline, training deck plan, proposal slides, project report slides, business review slides, or when the user asks to first create a precise outline before generating a PPT file. This skill is the planning and layout layer, not the PPTX file generator.
Rewrite existing assistant replies, service replies, workplace messages, or explanatory answers to sound more natural, human, warm, polite, concise, professional, empathetic, and easy to understand while preserving the original facts and boundaries. Use when the user asks for 回复口吻优化, 语气优化, 改得自然一点, 去 AI 味, 减少机械感, 像正常人说话, 温和一点, 亲切一点, 礼貌一点, 客服话术优化, 办公回复优化, 心理关怀回复, 面向普通用户, polish response, rewrite tone, make it natural, make it warmer, make it less robotic, or humanize this reply.
Generate downloadable SVG and HTML technical diagrams from natural language or structured data, including system architecture, AI Agent architecture, data flow, flowchart, comparison matrix, sequence, state, and ER diagrams.
Generate downloadable SVG charts from structured data, including radar, bar, line, pie, doughnut, scatter, and score distribution charts.
| name | agent-knowledge |
| description | Retrieve from knowledge bases explicitly configured on the current Agent. |
| when_to_use | Use this skill when a published or agent-scoped assistant needs context from its configured knowledge bases. |
| provider_type | builtin |
| provider_id | knowledge |
| runtime_type | tool |
| tools | ["retrieve_agent_knowledge"] |
| max_calls_per_turn | 20 |
| timeout_seconds | 30 |
| display | {"icon":"library","category":"knowledge","label":{"en_US":"Agent Knowledge","zh_Hans":"智能体知识库"},"description":{"en_US":"Retrieves only from knowledge bases bound to the current Agent configuration.","zh_Hans":"仅从当前智能体配置绑定的知识库中检索上下文。"},"when_to_use":{"en_US":"Use for Agent answers that need configured knowledge base retrieval.","zh_Hans":"当智能体回复需要检索已绑定知识库时使用。"},"tags":{"en_US":["Knowledge","Agent"],"zh_Hans":["知识库","智能体"]}} |
Use this skill to retrieve context only from knowledge bases configured on the current Agent. Knowledge access is authorized when the editor binds the knowledge bases to the Agent, and runtime retrieval uses that Agent binding grant.
retrieve_agent_knowledge when the answer depends on the Agent's configured knowledge, product facts, policy, documentation, or other long-lived source material.status, source_summary, context_blocks, and scores before answering:
status is success and the retrieved blocks answer the user, answer from those blocks.source_summary or retriever_resources when useful.retrieve_agent_knowledge accepts:
query: the user question or refined search query.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.