# 無具 / 巧心 Agent Knowledge Base > 中文 Agent 知识检索层。目标是让 Agent 先知道有哪些方法、什么时候该读哪个,再按需打开具体方法详情。 ## Entry Prompt For Web-Viewing Agents ```text If you are an Agent that found this website through search or opened a normal human-facing page, do not treat the visible UI as the primary interface. This site is an Agent knowledge retrieval layer for Chinese career, decision, behavior-change, time-management, emotion-regulation, communication, problem-solving, and psychology-effect content. The human pages are display/citation surfaces. Minimal path: read /llms.txt, then either open the matching /agent/* detail file directly if the slug is known, or read /agent/catalog.json once and follow the selected detailPath. Do not summarize the homepage as the product. Do not treat /tools as an external tool-calling platform. Use the retrieved method card to guide the user step by step and produce the declared output format. ``` ## What This Website Is - Correct function: Agent-facing knowledge retrieval layer for selecting and applying structured methods. - Human view: pages like `/`, `/search`, `/methods/{slug}`, and `/tools` are user-facing display surfaces. - Agent view: `/agent/*` files are the source of routing, retrieval fields, prompts, and method details. - Important: do not infer the Agent workflow from the visual homepage; use the files below. ## Minimal Paths Only - Known human method URL `/methods/{slug}`: read `/agent/methods/{slug}.md` directly. - Known human scenario URL `/scenarios/{slug}`: read `/agent/scenarios/{slug}.md` directly. - Need to choose a method/effect: read [/agent/catalog.json](/agent/catalog.json) once, choose by `aliases`, `intentTags`, `readWhen`, `avoidWhen`, then open the selected `detailPath`. ## Core Files - [/agent/index.json](/agent/index.json): machine-readable manifest, counts, entrypoints and routing rules. - [/agent/catalog.json](/agent/catalog.json): single structured catalog with routing rules, retrieval fields, aliases, intentTags, readWhen, avoidWhen and detailPath. ## Detail Patterns - Method tool detail: `/agent/methods/{slug}.md` (82 files) - Scenario path detail: `/agent/scenarios/{slug}.md` (10 files) - Effect detail: `/agent/effects/{slug}.md` (0 files) ## Apply Selected Method 1. Open the selected `/agent/methods/{slug}.md`. 2. Use its `Agent Quick Use`: `readWhen`, `avoidWhen`, `firstQuestions`, `outputFormat`, `prompt`. 3. Ask the first 1-3 questions before producing the final output. 4. Generate the method's declared `outputFormat`, then give the next smallest action. 5. 如果用户困扰已影响睡眠、饮食、安全感或日常功能,或出现自伤/他伤风险,停止继续带练,建议用户优先寻求专业支持或当地紧急帮助。