Interview Script · information
Ask about the last time.
Create a customer interview guide with neutral, past-behavior questions, a flexible timing plan and a structured note-taking template.
Harbor Tool Library · Fictional discovery guide
Walk me through the last time you borrowed a tool, from deciding you needed it to returning it.
What was the hardest part of that particular borrowing experience?
What time or money did you actually spend arranging the loan?
A selection of the twelve saved questions.
For this work
Researchers, founders and service teams preparing a learning conversation.
Not a participant recruiter, recorder, transcription service or proof of demand.
Bring this / receive this
| You provide | You receive |
|---|---|
| Research objective, decision, participant context, time available and authorized background material. | A Markdown interview script, probes, timing plan and note-taking template. |
Use it in your assistant
Extract the official source ZIP. Install the folder with your assistant’s supported Agent Skill method, or ask a file-capable assistant to read SKILL.md and apply it. Model access and any host file tools are required; subscription or usage fees may apply. Supplied-text work needs no additional service API key. Further-reading links are optional and their access is not bundled.
What was actually checked
8 source and saved-artifact checks passed in this controlled run. The results are bounded to this fictional case; no general quality percentage is inferred.
Original source and boundaries
Original instructions by Pawel Huryn from phuryn/pm-skills, pinned at 07b34f64ee828df08953f505b7f747aef85ae7c6. MIT license, copyright and full selected subtree are preserved. Packaging documentation and promotional artwork are additions; the core SKILL.md is unchanged.
- A prepared guide is not a completed interview or validated demand.
- The source includes a hypothetical magic-wand prompt; this controlled guide chooses past-behavior questions instead.
- Interview quality, interviewer listening ratio and real participant consent were not measured.
Only a source-guided Codex preparation and bounded Python artifact checks are recorded. No clean native-client installation, Claude run, real participant research or business outcome is claimed.