技能列表

Convert meeting transcripts into evidence-linked decisions and action drafts, then block incomplete tasks before they are created in Feishu, Notion, Jira, or another task system. Use when a user asks to turn meeting notes into owners, deadlines, follow-ups, or an auditable meeting action register.

Evaluate RAG retrieval runs against a versioned golden query set using Hit@K, Recall@K, MRR, nDCG and no-answer correctness; compare a candidate run with a baseline, block critical regressions, and create a source-level repair queue. Use before changing chunking, embeddings, metadata filters, hybrid search, rerankers or knowledge-base versions.

Prioritize authorized B2B website, event, CRM, marketplace, or partner inquiries with configurable ICP rules, deterministic evidence scoring, duplicate detection, missing-data flags, consent holds, and reviewable next-best actions. Use when sales teams need a ranked queue without scraping personal data, inventing enrichment, or automatically contacting prospects.

Triage authorized customer-support messages with privacy redaction, prompt-injection checks, urgency and emotion routing, approved knowledge-base retrieval, grounded response drafts, knowledge-gap detection, and context-rich human handoff. Use for email, website chat, helpdesk, enterprise WeChat, Feishu, or webhook support workflows where replies must remain reviewable and high-risk cases need early human intervention.

Review authorized resumes against explicit job requirements with privacy redaction, quoted evidence, transparent scoring, missing-skill checks, interview-question generation, and mandatory human approval. Use when an HR team needs a repeatable first-pass review queue but must not let an AI make final hiring, rejection, compensation, or employment decisions.

Review authorized resumes against explicit job requirements with privacy redaction, quoted evidence, transparent scoring, missing-skill checks, interview-question generation, and mandatory human approval. Use when an HR team needs a repeatable first-pass review queue but must not let an AI make final hiring, rejection, compensation, or employment decisions.

Coordinate a platform-neutral team of AI agents to turn one content brief into a researched, cited, drafted, reviewed, and publish-ready content package. Use for article, social post, newsletter, short-video script, campaign, or knowledge-content workflows that need role separation, handoffs, source traceability, duplicate-claim checks, privacy scanning, and human approval before publication.

Convert user-authorized Markdown or plain-text source material into a real editable PPTX, speaker-notes Markdown, and a per-slide source map. Use when a user needs a report, proposal, training deck, project update, or presentation generated from existing material without losing traceability.

将商品视觉简报与 MiniMax 媒体素材清单转换为响应式营销落地页的结构、文案、组件、动效和验收规范，并配合官方 frontend-dev Skill 完成页面实现与本地质量检查。用于商品落地页、活动页、产品页和 AI 生成素材驱动的营销网站建设。

将已核验的商品视觉简报转换成 MiniMax 图片、视频和配音素材的提示词、参数、保存路径与安全执行清单，并可在用户明确确认后通过 mmx CLI 生成本地素材。用于商品营销图、6秒视频、旁白音频、落地页媒体包和多模态内容生产。

使用 MiniMax vision-analysis 或 mmx vision describe 分析已授权的商品图片、页面截图和包装图，提取可核验事实、视觉风格、可用卖点、风险表述和素材缺口，输出可被后续 MiniMax 媒体生成与落地页工作流直接读取的商品视觉简报。用于商品视觉分析、营销素材准备、落地页改版和短视频素材策划。

Analyze user-authorized task logs, notes, CSV, JSON, JSONL, Markdown, or text files to find repeated work, score automation opportunities, and produce an evidence-based workflow specification. Use when a user wants to discover which recurring tasks should become an Agent Skill, scheduled automation, n8n/Dify workflow, script, or SOP.