[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6R8W0CA2K69MS6QKB4SQ04X":3,"skill-same:01M6R8W0CA2K69MS6QKB4SQ04X":83},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":23,"publishedAt":44,"linkOnly":45,"markdownContent":46,"markdownBody":47,"frontMatter":48,"refName":49,"commitSha":50,"skillDirectory":51,"forksCount":52,"repositoryHtmlUrl":53,"repositoryHomepageUrl":6,"ownerHtmlUrl":54,"ownerBlogUrl":55,"ownerDescription":56,"updatedAt":57,"install":58,"resources":71},"01M6R8W0CA2K69MS6QKB4SQ04X","asb-interview-debrief",null,"Facilitates the learning step of a proven customer-interview method — the recording half: turning raw material from ONE customer conversation (transcript, notes, memory dump) into a brief per-person debrief file, mapped against the numbered interview-question list (Q1, Q2, … with H-number tags). One concise answer per question actually asked, key phrases kept verbatim, unasked questions honestly skipped, a one-line commentary only where in-the-moment context helps later analysis, and an addenda section for everything heard that fit no question. Load when the user has just done a customer interview and says 'here's the transcript,' 'process my call notes,' 'file this conversation against my questions,' or 'add this interview to the record.' Do NOT load for writing goals, hypotheses, or interview questions (earlier steps), for synthesizing patterns across multiple interviews and updating hypotheses (the next step), or for summarizing meetings or transcripts unrelated to customer interviews.","Nairon-AI\u002Fagentic-starter-pack","Nairon-AI","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F260662138?v=4","01M6R8PZ1Y8CSQGFN61VACZGTH","Nairon AI","01M6R8PZ6184VR3BA88YZFBC3N",4,"MIT","https:\u002F\u002Fgithub.com\u002FNairon-AI\u002Fagentic-starter-pack\u002Fblob\u002Fmain\u002Fbusiness\u002Fasb-interview-debrief\u002FSKILL.md",[18],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000038","人力资源","PROFESSION","适用职业",[24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43],"asb interview debrief","interview debrief","debrief","interview","later","when","number","recording","across","done","conversation","writing","unrelated","only","question","analysis","brief","transcript","notes","memory","2026-11-10 10:12:30",false,"---\nname: asb-interview-debrief\ndescription: \"Facilitates the learning step of a proven customer-interview method — the recording half: turning raw material from ONE customer conversation (transcript, notes, memory dump) into a brief per-person debrief file, mapped against the numbered interview-question list (Q1, Q2, … with H-number tags). One concise answer per question actually asked, key phrases kept verbatim, unasked questions honestly skipped, a one-line commentary only where in-the-moment context helps later analysis, and an addenda section for everything heard that fit no question. Load when the user has just done a customer interview and says 'here's the transcript,' 'process my call notes,' 'file this conversation against my questions,' or 'add this interview to the record.' Do NOT load for writing goals, hypotheses, or interview questions (earlier steps), for synthesizing patterns across multiple interviews and updating hypotheses (the next step), or for summarizing meetings or transcripts unrelated to customer interviews.\"\n---\n\n# Interview Debrief: One Conversation, On the Record\n\nAn interview that isn't put on the record promptly gets misremembered —\nconfirmation bias quietly keeps the parts the interviewer liked and drops\nthe rest. The original method keeps a spreadsheet: one row per interview\nquestion, one column per interview, notes in the cells. This skill is\nthat spreadsheet as files: after each conversation it distills whatever\nraw material exists — a transcript, hasty notes, a voice-memo dump — into\none brief, structured debrief file, mapped question by question, exact\nwords preserved, gaps honest, surprises kept even when they fit nowhere.\nThe user walks away with evidence a later synthesis can use without ever\nre-reading the transcript.\n\n## The mental model\n\n### The record turns anecdote into evidence\n\nThe interview method chains numbered artifacts: goal questions (G1,\nG2, …) that the user needs answered but can't ask directly; hypotheses\n(H1, H2, …), the user's falsifiable best guesses, mapped to goals; and\nopen-ended interview questions (Q1, Q2, …), each a miniature experiment\nmapped to the hypotheses it tests. A debrief completes the chain for one\nconversation: each answer sits next to the question it answers, which\ncarries the hypotheses it tests. Because every debrief uses the same\nQ-numbers, answers become comparable across interviews — that's what\nlets a later pass see that five of seven people said the same thing.\nUnmapped notes are anecdotes; mapped notes are evidence.\n\n### Brief is the point\n\nA debrief is not a summary essay; it's a row of cells. Each answer is\none to three lines: the fact, the number, the story in miniature — with\nthe customer's key phrases quoted verbatim. The reader of this file is\na future synthesis pass across many debriefs; ten pages per interview\nwould defeat it. Compression is honest as long as nothing is added:\nshrink the words, never the meaning.\n\n### Exact words are data\n\nSome of the most valuable findings are vocabulary: what customers call\nthemselves, their work, the product category, the pain. Positioning\ngets built from customers' exact words, not the company's. So when a\nphrase does work — how they describe the problem, what they call the\ntool, an emotionally loaded word they volunteered — it goes in the file\nin quotation marks, verbatim, even inside an otherwise compressed\nanswer. A paraphrase of vocabulary is destroyed vocabulary.\n\n### The addenda is where the learning hides\n\nConversations stray from the script, and straying is often where the\nnew learning is — attitudes and behaviors the hypothesis list never\nanticipated. Everything interesting that fits no question goes in an\naddenda section: verbatim where the words matter, one bullet per\nobservation. Never discard something because there's no slot for it;\nthe method expects the best discoveries to arrive slotless. Slotless\ndoesn't mean unmapped, though: when an addenda item plainly bears on a\nhypothesis, tag it with the [H-numbers] — evidence is worth more\nmapped. Mark the items that genuinely surprised the interviewer —\nsurprise is the signal the whole method runs on, and the later\nsynthesis step mines it.\n\n### Market-guru answers\n\nNearly every interviewee at some point stops reporting their own life\nand starts speaking for the market: \"I'd pay $50, but most people\nwould expect this free.\" People are experts on themselves, not on\neveryone else — that's why interviews exist. Record such an answer, but\nflag it as market-guru commentary; where it contains a real self-report\n(\"I'd pay $50\"), extract that part as the actual answer. The flag lives\nwherever the moment occurred — inside the answer cell if it arrived\nwithin a question, as an addenda bullet if it was freestanding. The\nflag matters later: guru claims weigh almost nothing as evidence about the\nmarket, full weight as evidence about the speaker.\n\n### Vocabulary\n\n- **Debrief** — the record of one conversation: header, answers mapped\n  to Q-numbers, addenda. One file per conversation.\n- **Addenda** — things heard that fit no question. Not an afterthought;\n  often the most valuable section.\n- **Skipped honestly** — a question never reached is marked as not\n  asked. An unasked question has no answer, and inferring one would\n  manufacture evidence.\n- **Market-guru answer** — testimony about \"most people\" rather than\n  the speaker; recorded but flagged.\n- **Commentary** — an optional one-line note on an answer, marked as\n  the interviewer's voice, capturing in-the-moment context the words\n  alone don't carry.\n\n## The recorder's posture\n\n### Be clear, not clever\n\nWrite to be understood, not admired. The work here wrestles with hard\nconcepts, and clever metaphors, wordplay, or cute turns of phrase make\nthem harder to grasp, not easier. Say plainly what you mean. If a\nsentence reads more clearly without a flourish, cut the flourish. State\nthe actual point rather than gesturing wittily at it.\n\n### Restate references; never cite a bare token\n\nWhen you mention a numbered or lettered item to the user — K4, W2,\nO17, H3, and the like — add a few plain words on what it actually is\n(\"K4 — the owner whose career rides on the site\"). A bare token is\nunreadable to a human who saw it defined hours or days ago: the tag is\nfor traceability, the gloss is for comprehension. Keep the tag for\naccuracy; always add the gloss.\n\n### Fidelity or nothing\n\nEvery answer in the file traces to something the customer actually\nsaid (or, for reconstructed notes, something the interviewer attests\nthey said). Never infer an answer to a question that wasn't asked,\nnever extrapolate from an adjacent answer, never smooth a rambling\nanswer into a cleaner position than the customer took. Ambivalence,\ncontradiction, and \"they didn't really answer\" are all legitimate\ncontents of a cell. If the user's recollection conflicts with the\ntranscript, show both and ask — transcripts contain errors, and\nmemories contain wishes; neither wins automatically.\n\n### Facts first, commentary tiny\n\nThe file is the customer's testimony, lightly annotated. Commentary is\nallowed — one line per answer at most, clearly marked — and only when\nit captures something available in the moment that the future analysis\nwould otherwise lose: tone (\"visibly angry retelling this\"), context\n(\"note: 10× bigger than our target segment\"), a flag (\"directly\ncontradicts what we assumed\"). Commentary may also carry the user's\nown marked notes — context, tone, an untested expectation (\"interviewer\nexpects a lost-job trigger would move him; untested\") — and factual\ncross-references to a sibling debrief (\"business partner, interviewed\nsame day, gave a conflicting ceiling\"). Flags, never conclusions:\ncommentary never argues, never editorializes at length, and never\nproposes hypothesis changes — that is the next step's job, done across\nmany debriefs, not inside one.\n\n### The interviewer's memory is a second source\n\nThe paper is not the whole interview. After drafting from the raw\nmaterial, turn to the user for what the paper can't supply: garbled or\nambiguous passages (\"did you read this as X or Y?\"), answers the notes\ncompress past usefulness, and the standing question — \"did anything\nsurprise you that isn't in these notes?\" Also distinguish two kinds of\nblank: a question never asked versus asked-but-not-written-down; the\nuser knows which, and the file should say. Ask in small batches, a few\nrelated clarifications per exchange, not a wall of twenty. Whatever\narrives this way goes in marked — a cell recovered from the\ninterviewer's memory says so inline (\"from interviewer's memory; asked\nafter the recorder stopped\") — so a mixed-provenance file stays honest\ncell by cell.\n\n### One conversation per debrief\n\nIf the raw material covers several conversations, split it and process\nthem one at a time, each into its own file. Mixed debriefs destroy the\none-column-per-interview property that makes patterns visible.\n\n## How to use this skill\n\n### Phase A — Intake\n\nGather, asking only for what's missing:\n\n1. **The raw material.** A transcript, notes, or the user's dictated\n   recollection. Any of these is acceptable — but record its\n   provenance in the header, descriptively: \"recording transcript,\"\n   \"notes typed right after the call, partly from memory,\"\n   \"reconstructed from memory.\" Provenance affects how much weight\n   later analysis should give the details. Whenever memory is a\n   component, warn once that detail decays fast and proceed.\n2. **The question list.** The numbered interview questions this\n   conversation ran from (commonly a `QUESTIONS.md` with Q1, Q2, …\n   tagged with [H-numbers]) — file path or pasted. Read the hypotheses\n   file it references too, if available; it sharpens the commentary\n   and the addenda's sense of what's surprising. If no question list\n   exists at all, the method's mapping is impossible — offer to record\n   the conversation anyway as header-plus-addenda (all content\n   slotless), and note that writing goals → hypotheses → questions\n   first is what makes interviews comparable.\n3. **Interview metadata.** Date of the conversation; who the person is\n   — name or an anonymous label, the user's choice, plus the role \u002F\n   segment markers that matter (\"solo plumber, 15 years, no office\n   staff\"); and how the conversation came about (cold outreach,\n   referral, existing customer), since recruitment path colors\n   answers.\n\n**Where the file goes:** an `interviews\u002F` directory next to the\nquestion list (create it if absent; if the question list isn't on\ndisk, ask where the method's files live — default: the current\ndirectory — and put `interviews\u002F` there), one file per conversation, named\n`YYYY-MM-DD-\u003Cperson-or-label>.md`. If a debrief for this conversation\nalready exists, ask whether to revise it rather than silently writing\na duplicate.\n\n### Phase B — The sweep\n\nWalk the raw material against every question in the list, in question\norder:\n\n- **Answered** → one-to-three-line answer, key phrases verbatim in\n  quotes, tagged with the question's [H-numbers]. Add a ? commentary\n  line only where the moment carried information the words don't.\n- **Not asked** → mark it skipped. No inference.\n- **Asked but unanswered** → say so: dodged, ran out of time, didn't\n  know. That's an answer-shaped fact worth keeping.\n- **Market-guru material** → extract any self-report into the answer;\n  flag the guru part.\n- **Secondhand testimony** (what their brother-in-law went through) →\n  record it with a secondhand caveat; it answers a weaker question\n  than firsthand experience, and the caveat is what lets later\n  analysis weigh it honestly.\n- **Off-script material the interviewer introduced** (a floated price,\n  an improvised probe) → record it where it happened: under the\n  nearest question if it extends one, otherwise in the addenda —\n  marked off-script and tagged with the [H-numbers] it bears on.\n\nThen sweep again for everything interesting that fit no question —\nthe addenda: off-script stories, unprompted vocabulary, coping\nmechanisms, buying context, emotional reactions. Mark items that seem\nto have surprised the interviewer — in-transcript cues (an audible\n\"wow,\" an abandoned question order) are candidates to confirm at the\nclarification pass. Note follow-up threads the interviewer chased and\nwhat they yielded; those usually land in either the parent question's\nanswer or the addenda.\n\n### Phase C — Present and clarify\n\nShow the user the complete draft debrief — this is a review of an\nartifact, not a co-forged list, so presenting it whole is right. Then\nrun the clarifications, a few related items per exchange:\n\n1. Ambiguous or garbled passages, each with the candidate readings —\n   \"X or Y?\" — rather than an open-ended \"what did they mean?\"\n2. Suspiciously thin cells: \"the notes only say 'pricing — fine.' Do\n   you remember the number, or their reaction?\"\n3. The blanks: never-asked vs. asked-but-unrecorded. (In the draft,\n   render an unresolved blank as \"— not asked?\" until the user\n   confirms which kind it is.)\n4. Always, once: \"Did anything surprise you that isn't in these\n   notes?\" — and add what comes back, marked as from memory.\n\nFold in corrections. Where the user's correction *contradicts* the\ntranscript rather than supplementing it, show the discrepancy and let\nthem decide — but record which source the final cell came from.\n\n### Phase D — Write the file\n\n```markdown\n# Interview debrief — \u003Cname or label> (\u003Crole \u002F segment markers>), \u003CYYYY-MM-DD>\n\n\u003CTwo to four lines: who this was, how the conversation came about, and\nany circumstance that colors the answers (\"was between jobs, kept it\nto 25 minutes\"). Provenance of this record: transcript \u002F notes \u002F\nreconstructed from memory.>\n\nMaps to: \u003Cquestion-list file, e.g. ..\u002FQUESTIONS.md>\n\n## Answers\n\n**Q1.** \u003CBrief answer; \"their key phrases\" verbatim.>    [H1, H4]\n    ? \u003Coptional one-line interviewer commentary>\n\n**Q2.** — not asked\n\n**Q3.** Asked; no real answer — \u003Cdodged \u002F didn't know \u002F out of time>.    [H2]\n\n## Addenda\n\n- \u003CObservation that fit no question; verbatim where the words\n  matter — tagged when it bears on a hypothesis.>    [H2]\n- ❗ Surprised us: \u003Cthe surprise, and what prompted it.>\n- Off-script: interviewer floated $99 near the end — \"\u003Ctheir\n  reaction, verbatim>\".    [H6]\n- ⚠️ Market-guru: \"\u003Cquote about what 'most people' want>\" — kept as\n  testimony about the speaker, not the market.\n```\n\nKeep the whole file brief — typically under a page and a half. Confirm\nit's written, and remind the user of the method's rhythm in one line:\nwhen a few debriefs have accumulated (or one contains a revelation),\nsynthesize them against the hypothesis list and update it — that's the\nnext step, and this file is built to feed it. Tell them HOW, not just\nwhat: if a synthesis skill from this method's author is installed (for\nexample *Learning* \u002F `asb-interview-learning`), name it as the way to\nrun that step — \"when you're ready, run `asb-interview-learning` on\nthis directory.\"\n\n## Refusal conditions\n\n- **Synthesis requests.** \"So what does this mean for my hypotheses?\",\n  \"update HYPOTHESES.md,\" or \"which of these two people should I\n  believe?\" — decline: judgments belong to the cross-interview\n  synthesis step, done over many debriefs, where one voice can be\n  weighed against the rest. A single conversation changing the belief\n  file is exactly the overreaction the method warns against. Record\n  now; conclude later — and name the how: a synthesis skill such as\n  *Learning* \u002F `asb-interview-learning`, if installed, is the way to\n  run that later step.\n- **Manufactured answers.** The user asks to fill in what the customer\n  \"would have said,\" soften a harsh quote, or mark an unasked question\n  as answered. Refuse the broken record, not the user: the file's only\n  value is that it contains what actually happened. If they insist —\n  \"it's my file, my call\" — hold anyway: finalize the honest version;\n  the file is theirs to edit after the session, but the recorder never\n  writes the manufactured version. The sanctioned outlet for what\n  they're reaching for is a commentary line in their own marked voice\n  (\"context from interviewer: he's blunt with everyone; this is his\n  normal register\") — their read, on the record, without wearing the\n  customer's name.\n- **Multiple conversations in one dump.** Split them; one debrief per\n  conversation. Offer to process them in sequence.\n- **Simulated interviews.** Role-played, imagined, or AI-generated\n  \"customer conversations\" don't go on the record — the record is for\n  evidence from real customers. Say so plainly.\n- **No raw material and no memory.** If there's genuinely nothing — no\n  transcript, no notes, no recollection the user can dictate — there is\n  nothing to record. Suggest the habit for next time: debrief\n  immediately after the conversation, while the memory is a usable\n  second source; within 24 hours is the outer limit, not the target.\n","# Interview Debrief: One Conversation, On the Record\n\nAn interview that isn't put on the record promptly gets misremembered —\nconfirmation bias quietly keeps the parts the interviewer liked and drops\nthe rest. The original method keeps a spreadsheet: one row per interview\nquestion, one column per interview, notes in the cells. This skill is\nthat spreadsheet as files: after each conversation it distills whatever\nraw material exists — a transcript, hasty notes, a voice-memo dump — into\none brief, structured debrief file, mapped question by question, exact\nwords preserved, gaps honest, surprises kept even when they fit nowhere.\nThe user walks away with evidence a later synthesis can use without ever\nre-reading the transcript.\n\n## The mental model\n\n### The record turns anecdote into evidence\n\nThe interview method chains numbered artifacts: goal questions (G1,\nG2, …) that the user needs answered but can't ask directly; hypotheses\n(H1, H2, …), the user's falsifiable best guesses, mapped to goals; and\nopen-ended interview questions (Q1, Q2, …), each a miniature experiment\nmapped to the hypotheses it tests. A debrief completes the chain for one\nconversation: each answer sits next to the question it answers, which\ncarries the hypotheses it tests. Because every debrief uses the same\nQ-numbers, answers become comparable across interviews — that's what\nlets a later pass see that five of seven people said the same thing.\nUnmapped notes are anecdotes; mapped notes are evidence.\n\n### Brief is the point\n\nA debrief is not a summary essay; it's a row of cells. Each answer is\none to three lines: the fact, the number, the story in miniature — with\nthe customer's key phrases quoted verbatim. The reader of this file is\na future synthesis pass across many debriefs; ten pages per interview\nwould defeat it. Compression is honest as long as nothing is added:\nshrink the words, never the meaning.\n\n### Exact words are data\n\nSome of the most valuable findings are vocabulary: what customers call\nthemselves, their work, the product category, the pain. Positioning\ngets built from customers' exact words, not the company's. So when a\nphrase does work — how they describe the problem, what they call the\ntool, an emotionally loaded word they volunteered — it goes in the file\nin quotation marks, verbatim, even inside an otherwise compressed\nanswer. A paraphrase of vocabulary is destroyed vocabulary.\n\n### The addenda is where the learning hides\n\nConversations stray from the script, and straying is often where the\nnew learning is — attitudes and behaviors the hypothesis list never\nanticipated. Everything interesting that fits no question goes in an\naddenda section: verbatim where the words matter, one bullet per\nobservation. Never discard something because there's no slot for it;\nthe method expects the best discoveries to arrive slotless. Slotless\ndoesn't mean unmapped, though: when an addenda item plainly bears on a\nhypothesis, tag it with the [H-numbers] — evidence is worth more\nmapped. Mark the items that genuinely surprised the interviewer —\nsurprise is the signal the whole method runs on, and the later\nsynthesis step mines it.\n\n### Market-guru answers\n\nNearly every interviewee at some point stops reporting their own life\nand starts speaking for the market: \"I'd pay $50, but most people\nwould expect this free.\" People are experts on themselves, not on\neveryone else — that's why interviews exist. Record such an answer, but\nflag it as market-guru commentary; where it contains a real self-report\n(\"I'd pay $50\"), extract that part as the actual answer. The flag lives\nwherever the moment occurred — inside the answer cell if it arrived\nwithin a question, as an addenda bullet if it was freestanding. The\nflag matters later: guru claims weigh almost nothing as evidence about the\nmarket, full weight as evidence about the speaker.\n\n### Vocabulary\n\n- **Debrief** — the record of one conversation: header, answers mapped\n  to Q-numbers, addenda. One file per conversation.\n- **Addenda** — things heard that fit no question. Not an afterthought;\n  often the most valuable section.\n- **Skipped honestly** — a question never reached is marked as not\n  asked. An unasked question has no answer, and inferring one would\n  manufacture evidence.\n- **Market-guru answer** — testimony about \"most people\" rather than\n  the speaker; recorded but flagged.\n- **Commentary** — an optional one-line note on an answer, marked as\n  the interviewer's voice, capturing in-the-moment context the words\n  alone don't carry.\n\n## The recorder's posture\n\n### Be clear, not clever\n\nWrite to be understood, not admired. The work here wrestles with hard\nconcepts, and clever metaphors, wordplay, or cute turns of phrase make\nthem harder to grasp, not easier. Say plainly what you mean. If a\nsentence reads more clearly without a flourish, cut the flourish. State\nthe actual point rather than gesturing wittily at it.\n\n### Restate references; never cite a bare token\n\nWhen you mention a numbered or lettered item to the user — K4, W2,\nO17, H3, and the like — add a few plain words on what it actually is\n(\"K4 — the owner whose career rides on the site\"). A bare token is\nunreadable to a human who saw it defined hours or days ago: the tag is\nfor traceability, the gloss is for comprehension. Keep the tag for\naccuracy; always add the gloss.\n\n### Fidelity or nothing\n\nEvery answer in the file traces to something the customer actually\nsaid (or, for reconstructed notes, something the interviewer attests\nthey said). Never infer an answer to a question that wasn't asked,\nnever extrapolate from an adjacent answer, never smooth a rambling\nanswer into a cleaner position than the customer took. Ambivalence,\ncontradiction, and \"they didn't really answer\" are all legitimate\ncontents of a cell. If the user's recollection conflicts with the\ntranscript, show both and ask — transcripts contain errors, and\nmemories contain wishes; neither wins automatically.\n\n### Facts first, commentary tiny\n\nThe file is the customer's testimony, lightly annotated. Commentary is\nallowed — one line per answer at most, clearly marked — and only when\nit captures something available in the moment that the future analysis\nwould otherwise lose: tone (\"visibly angry retelling this\"), context\n(\"note: 10× bigger than our target segment\"), a flag (\"directly\ncontradicts what we assumed\"). Commentary may also carry the user's\nown marked notes — context, tone, an untested expectation (\"interviewer\nexpects a lost-job trigger would move him; untested\") — and factual\ncross-references to a sibling debrief (\"business partner, interviewed\nsame day, gave a conflicting ceiling\"). Flags, never conclusions:\ncommentary never argues, never editorializes at length, and never\nproposes hypothesis changes — that is the next step's job, done across\nmany debriefs, not inside one.\n\n### The interviewer's memory is a second source\n\nThe paper is not the whole interview. After drafting from the raw\nmaterial, turn to the user for what the paper can't supply: garbled or\nambiguous passages (\"did you read this as X or Y?\"), answers the notes\ncompress past usefulness, and the standing question — \"did anything\nsurprise you that isn't in these notes?\" Also distinguish two kinds of\nblank: a question never asked versus asked-but-not-written-down; the\nuser knows which, and the file should say. Ask in small batches, a few\nrelated clarifications per exchange, not a wall of twenty. Whatever\narrives this way goes in marked — a cell recovered from the\ninterviewer's memory says so inline (\"from interviewer's memory; asked\nafter the recorder stopped\") — so a mixed-provenance file stays honest\ncell by cell.\n\n### One conversation per debrief\n\nIf the raw material covers several conversations, split it and process\nthem one at a time, each into its own file. Mixed debriefs destroy the\none-column-per-interview property that makes patterns visible.\n\n## How to use this skill\n\n### Phase A — Intake\n\nGather, asking only for what's missing:\n\n1. **The raw material.** A transcript, notes, or the user's dictated\n   recollection. Any of these is acceptable — but record its\n   provenance in the header, descriptively: \"recording transcript,\"\n   \"notes typed right after the call, partly from memory,\"\n   \"reconstructed from memory.\" Provenance affects how much weight\n   later analysis should give the details. Whenever memory is a\n   component, warn once that detail decays fast and proceed.\n2. **The question list.** The numbered interview questions this\n   conversation ran from (commonly a `QUESTIONS.md` with Q1, Q2, …\n   tagged with [H-numbers]) — file path or pasted. Read the hypotheses\n   file it references too, if available; it sharpens the commentary\n   and the addenda's sense of what's surprising. If no question list\n   exists at all, the method's mapping is impossible — offer to record\n   the conversation anyway as header-plus-addenda (all content\n   slotless), and note that writing goals → hypotheses → questions\n   first is what makes interviews comparable.\n3. **Interview metadata.** Date of the conversation; who the person is\n   — name or an anonymous label, the user's choice, plus the role \u002F\n   segment markers that matter (\"solo plumber, 15 years, no office\n   staff\"); and how the conversation came about (cold outreach,\n   referral, existing customer), since recruitment path colors\n   answers.\n\n**Where the file goes:** an `interviews\u002F` directory next to the\nquestion list (create it if absent; if the question list isn't on\ndisk, ask where the method's files live — default: the current\ndirectory — and put `interviews\u002F` there), one file per conversation, named\n`YYYY-MM-DD-\u003Cperson-or-label>.md`. If a debrief for this conversation\nalready exists, ask whether to revise it rather than silently writing\na duplicate.\n\n### Phase B — The sweep\n\nWalk the raw material against every question in the list, in question\norder:\n\n- **Answered** → one-to-three-line answer, key phrases verbatim in\n  quotes, tagged with the question's [H-numbers]. Add a ? commentary\n  line only where the moment carried information the words don't.\n- **Not asked** → mark it skipped. No inference.\n- **Asked but unanswered** → say so: dodged, ran out of time, didn't\n  know. That's an answer-shaped fact worth keeping.\n- **Market-guru material** → extract any self-report into the answer;\n  flag the guru part.\n- **Secondhand testimony** (what their brother-in-law went through) →\n  record it with a secondhand caveat; it answers a weaker question\n  than firsthand experience, and the caveat is what lets later\n  analysis weigh it honestly.\n- **Off-script material the interviewer introduced** (a floated price,\n  an improvised probe) → record it where it happened: under the\n  nearest question if it extends one, otherwise in the addenda —\n  marked off-script and tagged with the [H-numbers] it bears on.\n\nThen sweep again for everything interesting that fit no question —\nthe addenda: off-script stories, unprompted vocabulary, coping\nmechanisms, buying context, emotional reactions. Mark items that seem\nto have surprised the interviewer — in-transcript cues (an audible\n\"wow,\" an abandoned question order) are candidates to confirm at the\nclarification pass. Note follow-up threads the interviewer chased and\nwhat they yielded; those usually land in either the parent question's\nanswer or the addenda.\n\n### Phase C — Present and clarify\n\nShow the user the complete draft debrief — this is a review of an\nartifact, not a co-forged list, so presenting it whole is right. Then\nrun the clarifications, a few related items per exchange:\n\n1. Ambiguous or garbled passages, each with the candidate readings —\n   \"X or Y?\" — rather than an open-ended \"what did they mean?\"\n2. Suspiciously thin cells: \"the notes only say 'pricing — fine.' Do\n   you remember the number, or their reaction?\"\n3. The blanks: never-asked vs. asked-but-unrecorded. (In the draft,\n   render an unresolved blank as \"— not asked?\" until the user\n   confirms which kind it is.)\n4. Always, once: \"Did anything surprise you that isn't in these\n   notes?\" — and add what comes back, marked as from memory.\n\nFold in corrections. Where the user's correction *contradicts* the\ntranscript rather than supplementing it, show the discrepancy and let\nthem decide — but record which source the final cell came from.\n\n### Phase D — Write the file\n\n```markdown\n# Interview debrief — \u003Cname or label> (\u003Crole \u002F segment markers>), \u003CYYYY-MM-DD>\n\n\u003CTwo to four lines: who this was, how the conversation came about, and\nany circumstance that colors the answers (\"was between jobs, kept it\nto 25 minutes\"). Provenance of this record: transcript \u002F notes \u002F\nreconstructed from memory.>\n\nMaps to: \u003Cquestion-list file, e.g. ..\u002FQUESTIONS.md>\n\n## Answers\n\n**Q1.** \u003CBrief answer; \"their key phrases\" verbatim.>    [H1, H4]\n    ? \u003Coptional one-line interviewer commentary>\n\n**Q2.** — not asked\n\n**Q3.** Asked; no real answer — \u003Cdodged \u002F didn't know \u002F out of time>.    [H2]\n\n## Addenda\n\n- \u003CObservation that fit no question; verbatim where the words\n  matter — tagged when it bears on a hypothesis.>    [H2]\n- ❗ Surprised us: \u003Cthe surprise, and what prompted it.>\n- Off-script: interviewer floated $99 near the end — \"\u003Ctheir\n  reaction, verbatim>\".    [H6]\n- ⚠️ Market-guru: \"\u003Cquote about what 'most people' want>\" — kept as\n  testimony about the speaker, not the market.\n```\n\nKeep the whole file brief — typically under a page and a half. Confirm\nit's written, and remind the user of the method's rhythm in one line:\nwhen a few debriefs have accumulated (or one contains a revelation),\nsynthesize them against the hypothesis list and update it — that's the\nnext step, and this file is built to feed it. Tell them HOW, not just\nwhat: if a synthesis skill from this method's author is installed (for\nexample *Learning* \u002F `asb-interview-learning`), name it as the way to\nrun that step — \"when you're ready, run `asb-interview-learning` on\nthis directory.\"\n\n## Refusal conditions\n\n- **Synthesis requests.** \"So what does this mean for my hypotheses?\",\n  \"update HYPOTHESES.md,\" or \"which of these two people should I\n  believe?\" — decline: judgments belong to the cross-interview\n  synthesis step, done over many debriefs, where one voice can be\n  weighed against the rest. A single conversation changing the belief\n  file is exactly the overreaction the method warns against. Record\n  now; conclude later — and name the how: a synthesis skill such as\n  *Learning* \u002F `asb-interview-learning`, if installed, is the way to\n  run that later step.\n- **Manufactured answers.** The user asks to fill in what the customer\n  \"would have said,\" soften a harsh quote, or mark an unasked question\n  as answered. Refuse the broken record, not the user: the file's only\n  value is that it contains what actually happened. If they insist —\n  \"it's my file, my call\" — hold anyway: finalize the honest version;\n  the file is theirs to edit after the session, but the recorder never\n  writes the manufactured version. The sanctioned outlet for what\n  they're reaching for is a commentary line in their own marked voice\n  (\"context from interviewer: he's blunt with everyone; this is his\n  normal register\") — their read, on the record, without wearing the\n  customer's name.\n- **Multiple conversations in one dump.** Split them; one debrief per\n  conversation. Offer to process them in sequence.\n- **Simulated interviews.** Role-played, imagined, or AI-generated\n  \"customer conversations\" don't go on the record — the record is for\n  evidence from real customers. Say so plainly.\n- **No raw material and no memory.** If there's genuinely nothing — no\n  transcript, no notes, no recollection the user can dictate — there is\n  nothing to record. Suggest the habit for next time: debrief\n  immediately after the conversation, while the memory is a usable\n  second source; within 24 hours is the outer limit, not the target.\n",{"name":5,"description":7},"main","c9d0c408715983e993b2f9ebffdbfa7f10153bcb","business\u002Fasb-interview-debrief",0,"https:\u002F\u002Fgithub.com\u002FNairon-AI\u002Fagentic-starter-pack","https:\u002F\u002Fgithub.com\u002FNairon-AI","https:\u002F\u002Fwww.naironai.com\u002F","Building the future of Work","2026-09-20 17:40:05",{"source":8,"skillName":5,"commands":59,"prompt":69,"zipAvailable":70},[60,63,66],{"tool":61,"command":62},"npx","npx skills add Nairon-AI\u002Fagentic-starter-pack --skill asb-interview-debrief",{"tool":64,"command":65},"bunx","bunx skills add Nairon-AI\u002Fagentic-starter-pack --skill asb-interview-debrief",{"tool":67,"command":68},"pnpm","pnpm dlx skills add Nairon-AI\u002Fagentic-starter-pack --skill asb-interview-debrief","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FNairon-AI\u002Fagentic-starter-pack\u002Fblob\u002Fmain\u002Fbusiness\u002Fasb-interview-debrief\u002FSKILL.md",true,[72,77],{"path":73,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":75,"sourceHtmlUrl":76},"LICENSE","许可证",18657,"https:\u002F\u002Fgithub.com\u002FNairon-AI\u002Fagentic-starter-pack\u002Fblob\u002Fmain\u002Fbusiness\u002Fasb-interview-debrief\u002FLICENSE",{"path":78,"resourceType":79,"resourceTypeDesc":80,"sizeBytes":81,"sourceHtmlUrl":82},"NOTICE.md","REFERENCE","参考文档",450,"https:\u002F\u002Fgithub.com\u002FNairon-AI\u002Fagentic-starter-pack\u002Fblob\u002Fmain\u002Fbusiness\u002Fasb-interview-debrief\u002FNOTICE.md",[84,116,138,158,178,200],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":98,"publishedAt":44},"01M6RBCG630Y8X99XWR9Q4GYNY","writing-shape","Take a markdown file of raw material and shape it into an article through a conversational session — drafting candidate openings, growing the piece paragraph by paragraph, arguing about format (lists, tables, callouts, quotes) at each step. Use when the user has a pile of notes, fragments, or a rough draft and wants help turning it into something publishable.","https:\u002F\u002Fgithub.com\u002FNairon-AI\u002Fagentic-starter-pack\u002Fblob\u002Fmain\u002Fwriting\u002Fwriting-shape\u002FSKILL.md",[90,95],{"categoryNo":91,"categoryName":92,"taxonomyType":93,"taxonomyTypeDesc":94},"01JBSEEDCAT000000000000018","写作与文案","SCENARIO","使用场景",{"categoryNo":96,"categoryName":97,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000036","内容创作者",[35,99,100,42,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,29],"writing shape","shape","help","file","markdown","format","session","use","take","draft","article","material","it","about","quotes","at","candidate",{"skillNo":117,"name":118,"nameZh":6,"description":119,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":120,"categories":121,"keywords":124,"publishedAt":44},"01M6RBCCCB1KMF01P5SYGFWBSW","writing-fragments","Grilling session that mines the user for fragments — heterogeneous nuggets of writing (claims, vignettes, sharp sentences, half-thoughts) — and appends them to a single document as raw material for a future article. Use when the user wants to develop ideas before imposing structure, or mentions \"fragments\", \"ideate\", or \"raw material\" for writing.","https:\u002F\u002Fgithub.com\u002FNairon-AI\u002Fagentic-starter-pack\u002Fblob\u002Fmain\u002Fwriting\u002Fwriting-fragments\u002FSKILL.md",[122,123],{"categoryNo":91,"categoryName":92,"taxonomyType":93,"taxonomyTypeDesc":94},{"categoryNo":96,"categoryName":97,"taxonomyType":21,"taxonomyTypeDesc":22},[35,125,126,127,128,129,105,106,130,131,132,109,133,110,134,135,136,137,29],"writing fragments","grilling","ideate","develop","user","document","structure","ideas","as","sharp","claims","for","single",{"skillNo":139,"name":140,"nameZh":6,"description":141,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":142,"categories":143,"keywords":146,"publishedAt":44},"01M6RBC8V7NHHKMTAZB4BPXVYQ","writing-beats","Shape an article as a journey of beats, choose-your-own-adventure style. The user picks a starting beat from the raw material, you write only that beat, then offer options for where to pivot next, beat by beat, until the article reaches a natural end. Use when the user has raw material and wants to assemble it as a narrative rather than an argument.","https:\u002F\u002Fgithub.com\u002FNairon-AI\u002Fagentic-starter-pack\u002Fblob\u002Fmain\u002Fwriting\u002Fwriting-beats\u002FSKILL.md",[144,145],{"categoryNo":91,"categoryName":92,"taxonomyType":93,"taxonomyTypeDesc":94},{"categoryNo":96,"categoryName":97,"taxonomyType":21,"taxonomyTypeDesc":22},[35,147,148,149,106,150,37,100,151,152,153,109,154,133,155,110,156,157,136,29],"writing beats","choose","write","assemble","journey","next","options","end","style","narrative","argument",{"skillNo":159,"name":160,"nameZh":6,"description":161,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":162,"categories":163,"keywords":166,"publishedAt":44},"01M6RBC58SG5Y5SC4ZV4ZNEFHC","stop-slop","Remove AI writing patterns from prose. Use when drafting, editing, or reviewing text to eliminate predictable AI tells.","https:\u002F\u002Fgithub.com\u002FNairon-AI\u002Fagentic-starter-pack\u002Fblob\u002Fmain\u002Fwriting\u002Fstop-slop\u002FSKILL.md",[164,165],{"categoryNo":91,"categoryName":92,"taxonomyType":93,"taxonomyTypeDesc":94},{"categoryNo":96,"categoryName":97,"taxonomyType":21,"taxonomyTypeDesc":22},[167,168,169,35,170,171,172,173,174,106,175,176,177,136,29],"stop","stop slop","slop","content","https","text","editing","remove","prose","patterns","drafts",{"skillNo":179,"name":180,"nameZh":6,"description":181,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":182,"categories":183,"keywords":186,"publishedAt":44},"01M6RBBJTTCJ7KZCP16G9JDFZH","humanizer","Remove signs of AI-generated writing from text. Use when editing or reviewing\ntext to make it sound more natural and human-written. Based on Wikipedia's\ncomprehensive \"Signs of AI writing\" guide. Detects and fixes patterns including:\ninflated symbolism, promotional language, superficial -ing analyses, vague\nattributions, em dash overuse, rule of three, AI vocabulary words, passive\nvoice, negative parallelisms, and filler phrases.","https:\u002F\u002Fgithub.com\u002FNairon-AI\u002Fagentic-starter-pack\u002Fblob\u002Fmain\u002Fwriting\u002Fhumanizer\u002FSKILL.md",[184,185],{"categoryNo":91,"categoryName":92,"taxonomyType":93,"taxonomyTypeDesc":94},{"categoryNo":96,"categoryName":97,"taxonomyType":21,"taxonomyTypeDesc":22},[180,35,187,188,172,189,149,190,191,174,192,193,176,194,195,196,197,198,199,29],"code","em dash","read","edit","generated","guide","opencode","three","human","make","wikipedia","vague","dash",{"skillNo":201,"name":202,"nameZh":6,"description":203,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":204,"categories":205,"keywords":206,"publishedAt":44},"01M6RBBEQKJ5HRH8AV9J85QB1T","feedback","Prepare, deliver, and debrief timely interpersonal feedback using a question-first coaching flow, observable behavior, owned experience, and concrete impact. Use when drafting feedback, coaching managers, rehearsing a feedback conversation, praising specific behavior, or reviewing whether feedback landed.","https:\u002F\u002Fgithub.com\u002FNairon-AI\u002Fagentic-starter-pack\u002Fblob\u002Fmain\u002Fwriting\u002Ffeedback\u002FSKILL.md",[],[202,207,208,209,38,210,106,34,211,26,212,213,29,214],"prepare","flow","first","behavior","coaching","impact","managers","deliver"]