Help us improve
Share bugs, ideas, or general feedback.
From startup-superpowers
Manages the full customer discovery interview lifecycle: drafting scripts, refining questions, managing script status, and analyzing transcripts to extract insights and update hypotheses.
npx claudepluginhub sergeigorbatiuk/startup-superpowers --plugin startup-superpowersHow this skill is triggered — by the user, by Claude, or both
Slash command
/startup-superpowers:interviewsThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Help the founder run the full customer discovery interview loop — from preparing a script to analyzing what was said afterward.
Generates customer interview scripts with JTBD probing questions and Mom Test principles for user research and discovery planning.
Design and conduct structured customer interviews to uncover problems, motivations, and decision-making.
Share bugs, ideas, or general feedback.
Help the founder run the full customer discovery interview loop — from preparing a script to analyzing what was said afterward.
This skill covers two phases:
An interview script is a reusable guide, not a teleprompter: it anchors the conversation around the assumptions being tested while leaving room for the founder to follow threads that open up. Most founders need one script for their primary segment; occasionally a second script for an adjacent segment.
Read startup/core.md to load project context (name, seed description, and all fields under ## Core — especially Audience).
Check if startup/interview-scripts/ contains any .md files.
Scaffold the folders if they don't exist yet:
mkdir -p startup/interview-scripts
mkdir -p startup/interviews/transcripts
Load and understand them for context. Infer intent from the conversation — don't mechanically ask "what do you want to do?" If the founder is:
draft — still being shaped, not yet usedready — actively in use for interviewsretired — superseded or no longer in rotationWhen adding or updating scripts, follow the file conventions:
status (draft, ready, retired), length_minutes (number — 15/30/45/60), and target_persona (one-line segment descriptor)## Target Persona — 2–4 sentences describing who this script is for and why## Opening — what the founder says at the start: purpose, consent to record, framing## Core Questions — numbered questions with optional indented probes## Closing — wrap-up, referrals ask, thank-you## Notes — facilitation tips for the founderSlug convention: lowercase the title, replace spaces and non-alphanumeric characters with hyphens, collapse multiple hyphens. "Freelance designers — invoice chasing" → freelance-designers-invoice-chasing.
Read before writing, propose before saving, get confirmation.
Check prerequisites:
core.md is missing or has no Audience or similar information that provides this information under ## Core: Mention that a defined target audience sharpens the script considerably — suggest filling that in first (via the whats-next skill or directly in core.md). Do not block: if the founder wants to proceed with a rough persona, proceed.
startup/hypotheses/ is empty or has fewer than 3 hypotheses: Mention that scripts work best when they're built around specific hypotheses to test — suggest invoking the hypotheses skill first. Do not block: if the founder insists on drafting a script now, proceed.
Prerequisites met (or founder insists): Load the reference file for the guided script-creation conversation:
.claude/skills/interviews/references/initial-interview-script.md
The reference file's instructions take over from this point.
After the founder runs an interview, they may bring back a transcript, paste text into chat, or simply recount a conversation from memory. In any of these cases, the workflow is the same: save the source material, dispatch the interview-analyst subagent, then dispatch the hypotheses-manager subagent, surface recommendations to the founder, and route confirmed hypothesis changes through the hypotheses skill.
Recognize this intent when the founder:
startup/interviews/transcripts/ (or similar)Load the Layer 2 reference file that owns this workflow end-to-end:
.claude/skills/interviews/references/transcript-analysis.md
The reference file's instructions take over from this point, including the three input branches (file / pasted / recollection), dispatching both subagents, summarizing to the founder, and routing confirmed edits through the hypotheses skill.
Do not extract statements, evaluate hypothesis state, or edit hypothesis files yourself — the reference file describes how to delegate that work to the two subagents.