ai-interview
Set up a QuestionPunk AI interview, in which an AI interviewer talks with each respondent by text chat or voice and asks follow-up questions, from a research goal or an existing discussion guide. Use for interview customers about why they cancelled, run voice interviews with our users, or turn my interview guide into an AI interview. Not for a fixed-question survey with no interviewer.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-14
- Stars
- ★ 1.6k
- Forks
- 491
01 前言 Front Matter
- name
- ai-interview
- description
- Set up a QuestionPunk AI interview, in which an AI interviewer talks with each respondent by text chat or voice and asks follow-up questions, from a research goal or an existing discussion guide. Use for interview customers about why they cancelled, run voice interviews with our users, or turn my interview guide into an AI interview. Not for a fixed-question survey with no interviewer.
03 SKILL.md 正文
Set up an AI interview
QuestionPunk's AI interviewer runs one conversation per respondent, in text or voice, from a goal you write. Your job is the brief it works from and the few fixed questions around it; the interviewer writes its own follow-ups during each interview.
Use the connected QuestionPunk tools. Host prefixes vary; resolve each tool by its unprefixed name, and search deferred or not-yet-loaded tools before deciding one is missing. Never invent a tool or parameter.
When you can't save to the account
If survey_create or question_create_bulk is missing after that search, QuestionPunk isn't connected in this chat. The usual cause is that it is installed but not signed in. Do not search a connector directory or use another vendor's tools. In your reply:
- Say how to connect. If a QuestionPunk
authenticatetool is listed, call it and give the sign-in link it returns. Otherwise: in Codex, runcodex mcp add questionpunk --url https://app.questionpunk.com/api/v1, thencodex mcp login questionpunk, and start a new chat; on another host, use its connector settings. - Write the interview plan in the chat (steps 1 and 2 below). If
survey_draft_linkis available, make a link with it: two to four short questions, then one question of typeai_interviewas the last question, whose text is the opening question the interviewer starts from. Give the link exactly as returned, and say it opens the draft as a guest, where QuestionPunk asks the opening question and then the AI interviewer follows up; the goal, topics to avoid, text or voice, and length are set in the editor after opening it, using the brief you wrote. Put only the interview in the link: no names, email addresses, customer lists or other confidential text from the conversation.
When the user says they connected, look for the tools again and carry on without asking for the request again.
Steps
- Get the goal: what decision the interviews inform and who is being interviewed. If the user gave a discussion guide or script, keep its intent and its must-ask topics. Ask one question only if the goal or audience is missing.
- Write the brief:
- Interview goal (
interviewGoal): what to learn, the topics to cover in rough order, and what a good answer explains (for example the moment they decided, what they tried first, what it cost them). Plain sentences, no more than about 150 words. - Greeting (
greetingInstruction): how the interviewer opens, in one or two sentences, including roughly how long it takes. - Avoid (
avoidTopics): what the interviewer must not ask or promise (pricing commitments, medical or legal advice, identifying details the study does not need). - A few fixed questions before the interview only when the analysis needs them (for example plan type or how long they have been a customer). Ask no demographics without a reason. A choice that needs "Other" gets the built-in one (
otherOption: truein a link,itemDetails.other_option: truewhen saving), not a plain "Other" option.
- Interview goal (
- Choose text or voice. Use voice when the user asked for voice or spoken answers; otherwise text. Length: 3 to 5 follow-up exchanges for a short check-in, 8 to 10 for a discovery interview; say which you chose.
- Call
survey_create(title at most 50 characters), thensurvey_getfor the language andversions.current. - Add everything with one
question_create_bulkcall: the fixed questions first, then oneanswer_type: "ai_moderated_interview"item whosetextis the opening question, withitemDetails{ "modality": "text" | "voice", "length": <exchanges> }andtextItemDetails{ "interviewGoal": ..., "greetingInstruction": ..., "avoidTopics": ... }. Useai_moderated_interview, not the olderai_intervieworai_audio_interview. If the interviewer should see earlier answers, follow the tool description forreferenceSources(voice reads onlyreferenceSources.perRespondent). - Read the survey back with
survey_getand check the interview question, its modality, length and brief are saved. Fix a setting withquestion_update, one write at a time with the version the last write returned. - Report the outline, text or voice, length, the brief in a few lines, the draft status and
editorUrl. Suggest the owner try it once with the editor's preview before sending it. Publish only if the user asked to publish or for a link to send; then callsurvey_publishand giveshareUrlexactly as returned.
Voice interviews and the AI interviewer may use the account's AI credits when respondents take them; say so when you report, without quoting prices you have not read from a tool. Creating a draft does not authorize publishing, sending invitations or paid recruitment. On authentication failure use the host's reconnect flow; never ask for a token in chat. If a create times out, look for the new survey with survey_list before creating another.
04 同仓库更多 Skills
worktable-review-with-annotations
Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.
worktable-find-and-synthesize-context
Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.
worktable-create-or-update-docs
Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.
worktable-create-or-manage-records
Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.
worktable-create-or-manage-interactive-html
Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.
worktable-collaborate-threads
Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.