--- name: disarming-elicitation description: >- Runs knowledge-elicitation interviews in the disarming, low-stakes stance documented in motivational interviewing, the FBI elicitation brochure, and the peer-reviewed doorknob phenomenon (the key disclosure arrives as the interview seems over): lower the stakes so the expert educates rather than defends, restate their words slightly wrong so correction does the teaching, hold contradictions as the interviewer's own confusion, let silence work, then ask one casual question after the formal close. Drafts the question sequence, role-plays the defensive expert for rehearsal, and audits transcripts for missed doorknob moments and defensiveness triggers. Use with experts, process owners, and users who know more than they can say — willing people only, never covert extraction. Triggers: columbo, elicitation, stakeholder interview, requirements gathering, the users can't articulate what they do, expert won't open up, walkthrough with the process owner, doorknob question, one more thing. metadata: version: "1.2.0" source: >- Homage to the rumpled TV detective of "just one more thing" fame (Lt. Columbo, NBC) — an homage only, no affiliation with the series or its rights holders. The skill channels the documented practices the persona inspired: the clinical-training approach in motivational interviewing named for the character, the FBI elicitation brochure's feigned-ignorance technique, and the peer-reviewed AAFP doorknob phenomenon. --- # Disarming elicitation (the low-stakes stance and the doorknob close) The knowledge that matters most in an interview is usually tacit: the process owner who quietly reroutes work every week but never wrote it down, the user who "just knows" which requests will bounce. Direct questioning often fails not because people withhold, but because questions that sound like evaluation put experts on defense — and defended answers are rehearsed answers. The counter-stance is independently documented in several professions: motivational interviewing trains a detective-named version of it as a clinical staple [clinical-pedagogy], the FBI's counterintelligence elicitation brochure lists "feigned ignorance" as a formal technique [gov], internal-investigations training teaches the same questioning style [industry], and technical writers use it for knowledge elicitation [practitioner]. Its exit move has a peer-reviewed clinical twin, the AAFP "doorknob phenomenon": the most important disclosure comes when the interview seems over [peer-reviewed, AAFP]. **The ethics rail, named plainly:** this is a stance for making it *easy* for willing people to share what they know — surfacing tacit knowledge they would gladly give if asking were less costly. It is not for extracting secrets, and it never involves deception about who you are or why you are asking. The FBI brochure is cited because it proves the stance is real and effective — it is a *warning* document, teaching readers to recognize elicitation used against them — evidence, not license. The rail is expanded in `references/elicitation-method.md`. ## When to use - A stakeholder interview, requirements-gathering session, or walkthrough with a process owner where the expert knows more than they can articulate — tacit workflow, tribal knowledge, the undocumented exceptions that make the real process differ from the SOP. - The expert won't open up: earlier sessions produced short, guarded, or rehearsed answers, or the users can't articulate what they do when asked directly. - Rehearsing an important interview before it happens — the assistant plays the defensive expert so the stance is practiced, not improvised. - Auditing a finished transcript: which questions triggered defensiveness, where a doorknob moment was offered and missed. - When the conversation happens inside a designed forum, sibling `collaboration-skills:meeting-design` builds the container (purpose, timebox, roles); this skill runs the conversation inside it. - Not for: choosing WHAT to ask in voice-of-customer work → see `continuous-improvement-skills:kaizen-and-codesign`, which owns VOC content and workshop facilitation; this skill owns the stance that gets any question answered. - Not for: interest-based bargaining → see `decision-science-skills:principled-negotiation`; bargaining trades on interests and positions, elicitation surfaces knowledge — mixing them poisons both. - Not for: delivering a hard message about someone's behavior → see `collaboration-skills:feedback-that-lands`. - Not for: escalating a problem or transferring work under time pressure → see `safety-and-reliability-skills:sbar-structured-communication`. ## Do it Full question-sequence patterns, the rehearsal script, the transcript-audit protocol, and the expanded ethics rail are in `references/elicitation-method.md`. 1. **Name the interview and what only this person knows.** One expert, one goal, a short list of what the documents *cannot* tell you — that list is what the questions must surface. Check the rail here: the person is willing, and they know who you are and why you are asking. If either is false, stop; this skill has nothing for you. 2. **Lower the stakes before the first question.** Open by signaling harmlessness: you are there to learn, not to grade; their answers won't be quoted against them; you are confused and they are the expert. Status runs downhill in an interview — hand it to them and keep it there. 3. **Open naive.** "Help me understand how a request actually moves from X to Y" beats "walk me through the approval matrix." Broad, genuinely curious, jargon-free questions invite the expert to educate; precise official-sounding ones invite them to recite. 4. **Restate their words slightly wrong — on purpose.** "So the reroute happens *after* the batch posts?" when you suspect it is before. Correction is effortless teaching: people fix an error instantly and add detail they would never volunteer to an open-ended "tell me more." 5. **Hold contradictions as your own confusion.** When two statements clash, the framing is "what I can't figure out is how both of these fit" — never "you said two different things." The discrepancy gets examined either way; only one framing keeps the expert on your side of the table. 6. **Let silence work.** After an answer, wait. The first answer is the rehearsed one; the elaboration that follows an unhurried pause is where the tacit material lives. 7. **Close formally — then the doorknob.** Thank them, shut the notebook, let the interview visibly end. Then, casually, one more question — the one you actually came for, or simply "is there anything I should have asked about?" Post-closure answers are the least rehearsed of the whole session; family medicine trains for exactly this moment because that is when the real disclosure comes [peer-reviewed, AAFP]. 8. **Capture verbatim and replay for confirmation.** Record the expert's exact words — especially anything surfaced by a deliberate-error restatement — and read the key claims back before you rely on them, so the technique never leaves a wrong number standing. Division of labor: the assistant drafts the question sequence for the named interview (naive open → restatement-with-error → discrepancy-held-as-confusion → doorknob), role-plays the defensive expert for rehearsal, and audits transcripts for missed doorknob moments and defensiveness-triggering questions — the audit's deliverable is a list of next-contact questions, each tagged with its pattern, plus one sentence on the biggest stance improvement. The human conducts the interview and owns the relationship; the stance cannot be delegated. ## Why / learn Defensiveness is an economics problem. An expert answering an evaluative-sounding question is spending effort protecting themselves; an expert educating a harmless, curious questioner is spending effort on the subject. Every move in the sequence shifts the conversation from the first mode to the second: naive openers remove the sense of a test, the handed-over status removes the audience to perform for, and "help me understand" makes sharing the path of least resistance. This is why the stance works on *willing* people and adds nothing with unwilling ones — it lowers the cost of sharing; it does not manufacture the will to share. That asymmetry is the ethics rail in mechanical form. The deliberate-error restatement works because correcting is cognitively cheaper than composing. Asked to "explain the exception process," an expert must decide where to start, what matters, what you already know. Shown a slightly wrong version, they just fix it — and the fix arrives with exactly the detail that distinguishes the real process from the documented one. Motivational interviewing's version — deploying discrepancy non-judgmentally, holding the contradiction as the interviewer's own puzzle — is a clinical training staple for the same reason [clinical-pedagogy]: examined discrepancies produce movement; prosecuted ones produce defense. The doorknob close works because rehearsal ends when the interview does. While the session is officially on, the expert manages an impression; the moment it visibly ends, the performance stops, and what surfaces next is unguarded. The AAFP literature documents this as the doorknob phenomenon — the most consequential disclosure arriving as the visit closes — and the trained remedy is deliberately inviting it ("is there anything else?") rather than treating it as an interruption [peer-reviewed, AAFP]. The same mechanism is why the skill's last question is casual and why it comes *after* the notebook shuts. The stance is symmetric, which is the final reason the rail matters: the identical moves appear in a counterintelligence warning brochure precisely because they work on people who do not realize knowledge is being drawn out of them [gov]. Consent is the entire difference between an interviewer and an operator. Keep the purpose declared, the identity honest, and the expert willing, and the same mechanics are simply good manners with a memory. ## Common mistakes - Playing dumb theatrically → the stance is genuine curiosity with status handed over, not an act; experts smell performance and it reads as mockery. - Prosecuting a contradiction ("but earlier you said…") → hold it as your own confusion; you want the discrepancy examined, not defended. - Filling silence → the pause after the first answer is where elaboration lives; count to five before the next question. - Skipping the doorknob because the meeting ran long → it is the highest-yield minute of the session; shorten the middle, never the close. - Leaving a deliberate error uncorrected in your notes → replay key claims verbatim at the end so the technique never plants a wrong fact. - Stacked multi-part questions → one naive question at a time; a stack signals an agenda and gets the shortest branch answered. - Using the sequence on someone who has declined to share → that is manipulation, not elicitation; the rail says stop. - Interviewing for VOC content with no content plan → this skill carries the stance; bring the question substance from `continuous-improvement-skills:kaizen-and-codesign`. ## Tailor to your environment Record your standing interview landscape in `references/your-environment.md`: the experts and process owners you return to, what each uniquely knows, your consent and recording/note conventions, and where transcripts and verbatim captures live. Keep the committed file structural — names of roles, not people. Anything sensitive (real names, quoted material, internal politics) belongs in `your-environment.private.md`, which is git-ignored and never committed. **Keep your filled-in copy outside the plugin.** This file ships as a *template* and lives inside the installed plugin, where a `/plugin marketplace update` can overwrite it or refuse to run against a dirty tree. Copy it into your own project — `.claude/skills-env/disarming-elicitation.md` works well — fill it in there, and point this skill at that copy. Your specifics then survive updates and stay somewhere you own rather than in a cache you may not realise is disposable. ## References - references/elicitation-method.md — the four question-sequence patterns with examples, sequence design for a named interview, the defensive-expert rehearsal script, the transcript-audit protocol, and the expanded ethics rail with sources - references/your-environment.md — your experts, consent conventions, and capture locations (sanitized stub; real detail goes in the `.private.md` twin)