Agent Skills
› mohitagw15856/pm-claude-skills
› faq-builder
faq-builder
GitHub从工单、聊天等真实数据中挖掘高频问题，使用用户原话构建FAQ。提供去重后的问题列表、结构化答案及持续更新的捕获循环机制，解决FAQ过时或难以查找的问题。
触发场景
创建产品或流程的FAQ
每周重复回答相同问题
将支持线程转化为文档
回答找不到答案的问题
安装
npx skills add mohitagw15856/pm-claude-skills --skill faq-builder -g -y
SKILL.md
Frontmatter
{
"name": "faq-builder",
"homepage": "https:\/\/mohitagw15856.github.io\/pm-claude-skills\/skill\/faq-builder.html",
"metadata": {
"openclaw": {
"emoji": "🧠"
}
},
"description": "Build an FAQ from the questions people actually ask — mined from tickets, chats, and repeated explanations, answered once and well, organized by the asker's words, and maintained by a capture loop instead of annual archaeology. Use when asked create an FAQ for this product\/process\/team, I answer the same questions weekly, turn our support threads into docs, or why does nobody find our answers. Produces the mined question list with frequencies, the answers in ask-language, the structure, and the capture loop."
}
FAQ Builder Skill
FAQs fail when they're written from the inside: questions the team wishes people asked ("What makes our approach unique?"), phrased in the team's vocabulary, frozen at launch. The working FAQ is mined — from tickets, chat searches, and the questions someone answers every week — phrased in the asker's words (findability is vocabulary-matching; nobody searches for "provisioning cadence" when their question is "how long until it works"), answered once and well, and kept alive by a capture loop: every newly-repeated question flows in, because the FAQ is a cache of answers and caches need writes.
What This Skill Produces
- The mined question list — real questions with rough frequencies and sources, deduplicated by intent
- The answers — direct-first (the answer, then the caveat), in ask-language, each with an owner
- The structure — ordered by frequency within asker-journey groups, question-phrased headings
- The capture loop — the answered-it-twice rule that keeps the FAQ current
Required Inputs
Ask for these if not provided:
- The question sources — tickets, chat logs, inbox searches, the team's "I keep explaining…" list; mining needs raw material, and the wish-list questions are explicitly not it
- The askers — customers, new hires, other teams? Their vocabulary (from the actual questions) becomes the headings' language
- The answer authority — who signs off that each answer is correct (an FAQ with confident wrong answers is worse than none); per-topic owners named
- Where it will live — the wiki, the docs site, the bot's knowledge base — findability mechanics differ, and answers should be written to be found there
Framework: The Builder Rules
- Mine, don't imagine: the question list comes from evidence — search the ticket system for repeated intents, grep the chat, ask each teammate for their three most-repeated explanations. Frequency ranks; a question asked weekly outranks a question asked once eloquently. Wish-questions ("Why are we the best?") are marketing, filed elsewhere.
- Dedupe by intent, keep the phrasings: "how do I reset my password" / "locked out" / "can't log in" are one answer with three findable phrasings — the heading takes the most common, the variants live in the body (or as aliases) so search hits all three.
- Answer-first, then nuance: the first sentence answers ("Yes — up to 5 seats on the standard plan."); caveats follow; links go last. Answers that open with background make the reader hunt, and hunting readers file tickets. Long answers get a link to the full doc rather than inlining it — the FAQ is the cache, not the library.
- Structure by the asker's journey, order by frequency: groups the asker recognizes (Getting started / Billing / When things break), most-asked first within each — not the org chart, not alphabetical. Headings are the questions verbatim, because headings are what search and scanning both hit.
- The capture loop is the maintenance: the rule — answered it twice outside the FAQ → it goes in this week — plus a quarterly prune of zero-traffic entries and a correctness pass by the owners. An FAQ without the loop decays into the confident-wrong-answers zone within two quarters, which is worse than its absence.
Output Format
FAQ: [audience/domain] — [N] questions, mined from [sources]
The Mined List
| Question (asker's words) | Frequency | Source | Variants |
|---|
The FAQ
[Group, in journey order]
[Question verbatim] — [Answer-first sentence. Caveat. Link.] (owner: [name])
The Capture Loop
[The answered-twice rule · where new candidates get logged · quarterly prune + correctness pass, calendared]
Quality Checks
- Every question traces to evidence with a frequency, not a wish
- Headings use the asker's vocabulary; variants are findable
- Every answer's first sentence answers
- Every answer has a named correctness owner
- The capture loop and prune are scheduled, not aspirational
Anti-Patterns
- Do not write questions nobody asked — the FAQ is a mirror of demand, not a brochure
- Do not phrase in internal jargon — findability is speaking the asker's language
- Do not open answers with background — answer, then explain
- Do not inline the library — cache the answer, link the depth
- Do not launch without the loop — a static FAQ is a snapshot aging into misinformation
版本历史
- 471c606 当前 2026-07-22 13:24
