# Asb Carol Observations > Facilitates the first step of a proven ideal-customer (ICP) method: gathering raw, honest, specific observations about what a company and product actually are — before any judgment about strengths or weaknesses. Walks the user through twelve unsparing question categories (what customers praise, the complaint with no defense, what separates your most profitable customers, and more) — or processes a team's write-storm notes one observation at a time — and records the results in OBSERVATIONS.md (numbered O1, O2, …), vivid and unevaluated. For a company operating online, it first scans public reviews and press into an External Research section that seeds it. Load when the user wants to figure out their ideal customer, take an honest look at their company, run a strengths-and-weaknesses exercise from scratch, or says 'who is our Carol' or 'what are we actually good at.' Do NOT load to classify observations into strengths and weaknesses (the next step), or for personal self-reflection unrelated to a company. - Skill: `asmartbear/asb-carol-observations` (Agent Skill) - Install (CLI): `npx skillmds@latest add asmartbear/asb-carol-observations` - Raw SKILL.md: https://api.skillmd.com/api/skills/asmartbear/asb-carol-observations/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Research & Search - Author: asmartbear (https://skillmd.com/u/asmartbear) - Updated: 2026-09-10 - Page: https://skillmd.com/skills/asmartbear/asb-carol-observations --- # Observations: The Raw Facts About Who You Actually Are Every ideal-customer definition is derived from an honest accounting of who the company actually is — but "write down our strengths and weaknesses" is an impossible instruction. Nobody knows where to start, everyone rationalizes, and whether something even IS a strength depends on who's asking. So this method starts a level lower: raw observations, gathered through twelve detailed question categories, recorded vividly and specifically, and deliberately NOT judged. Classification comes later; this step gets the truth on the record. ## The mental model ### Why observations before strengths Whether a fact is a strength or a weakness is in the eye of the beholder: "inexpensive" is a strength to price-conscious buyers and a weakness signal to serious ones; "one hundred features" is completeness to some and bloat to others. Judging too early also invites defense and rationalization, which kills honesty. So the procedure is: (1) generate raw facts — this skill; (2) distill facts into the attributes that matter; (3) classify each attribute — the next step. Trying to do all three at once produces the usual whiteboard of flattering vagueness. ### The outside-consultant posture Generate facts as if you were an outside consultant hired to reverse-engineer the company: What decisions has it made, even unintentionally? What must its strategy be, even if nobody wrote it down? You may observe behaviors and outcomes; you may NOT interrogate anyone about *why* they acted — asking why makes people unwittingly rationalize or mount a defense. This is discovery, not judgment. ### Public evidence is legitimate seed material When the company already operates and strangers already talk about it online, the outside-consultant's first move is to go read what they say. Public reviews, social posts, forum threads, news articles, and the company's own marketing are real artifacts — checkable by channel and quote — so gathering them up front is not fabrication; it is exactly the reverse-engineering this posture calls for. But seed is not verdict: a scraped review is a *candidate* observation, recorded in its own External Research section and then pressed and confirmed by the user (who alone has seen the private behaviors behind it) before it becomes a numbered observation. The research widens the aperture and pre-loads the walk; it never replaces it. For a company with no product or no public presence yet, there is nothing to scan — skip it. ### Don't evaluate, don't blame, don't act The standing rule of the whole session. "Customers post screenshots of support wait times" must not become "so we should hire more support people" — maybe fast support isn't strategic; maybe the fix is documentation, chat, product design, or a different market segment. Now is not the time. Equally: no observation is anyone's *fault*. The moment the session turns evaluative, people stop telling the truth. Ideas for features, marketing campaigns, or fixes WILL surface — good; they go in a side-list in the file, to be processed another time, and the session stays on task. ### Vivid and specific, or it isn't an observation "Support could be better" is a mood. "Customers post screenshots on Reddit of long ticket wait times" is an observation — it names a behavior, a channel, an artifact. Generic words (better, great, slow, many, some, quality) are interpreted differently by every reader and carry no evidence; every observation must contain the specific behavior, number, event, quote, or artifact that makes it checkable. "We love our customers" is banality; "any support rep can issue up to $500 in credits without approval" is a fact. When the user offers the mood, press for the incident behind it. Two common shapes to convert rather than reject: a **personality self-judgment** ("I'm bad at saying no") becomes the company behavior it produces ("3 of 9 clients got out-of-scope work last quarter, ~60 hours unbilled"); a **stakeholder's opinion** ("my cofounder thinks our pricing is the problem") is recordable as a who-said-what fact, filed under the category its *topic* belongs to, never as a verdict. ### The twelve categories Each category comes with its scope deliberately widened — the parentheticals matter, because they pre-empt the excuses people use to withhold: 1. **Undeniable comparative strength.** What do customers praise when they choose you despite your foibles? What does your product do so well even competitors admit it in their own sales calls? What can your team do that most teams cannot? *(Whether it's hard to copy or not, whether anyone seems to care or not.)* 2. **Consistent complaints.** What complaint have you heard so many times, in so many channels, you don't need data to know it's true? On what point does a competitor instantly win because you have no defense? What do people leave you over even while apologizing as they cancel? *(Whether it's smart to react or not, whether it's intrinsic or competitor-created.)* 3. **Proud of.** What about the product or team are you especially proud of — great workmanship, "who we are," hard to do, or just fun? *(Whether customers agree or not, whether there's data or not, whether it's an advantage or not.)* 4. **Head/tail differential.** What distinguishes your most profitable customers from the least profitable? Not what the best have in common — most of that is common to everyone; the insight is in the *differences*. *(Whether it was intentional or not, whether you think you should act on it or not.)* 5. **We wish / say we're great, but we're not.** What do you claim to be great at, not because it's true, but because customers wish it were true — and so do you? *(Whether or not it directly harms sales or retention.)* 6. **Customers advocate for.** When customers genuinely brag about you — social media, private meetings — what do they highlight? What would even a disgruntled ex-employee begrudgingly admit? *(Whether you think they're exaggerating or not, whether the majority would agree or not.)* 7. **Clear and present existential threats.** What is happening now, or has at least a 70% chance of happening within a few years, that would seriously disrupt the business — tank sales, trigger mass cancellations? *(Whether you can do anything about it or not, whether it's your fault or not. Hold the 70% bar — otherwise every company lists twenty theoretical worries.)* 8. **Organizational capabilities.** What does your org structure make easy or hard? Which decisions are fast and intelligent versus slow and overwrought? What do you execute with excellence, and what are you just not set up to do well? *(Whether it was deliberate or not, whether it matches best practice or not, whether changing it feels feasible or not.)* 9. **Technical architecture and capabilities.** What's easy to build because the architecture makes it easy, and what feels impossible no matter how important? What reliability, scale, or extensibility comes naturally, and what wobbles under load or needs constant heroics? Where does the architecture create real advantage, and where does it impose constraints you pretend are temporary but have been true for years? *(Whether the architecture is "good" or not, whether customers see it or not.)* 10. **Envy of / constantly losing sales to competitors.** What have you seen in other companies that you envy — things you feel in your bones are awesome but you don't have, especially if they cost you deals? *(Whether you should adopt it, or must face that it isn't who you are.)* 11. **Philosophy.** What do you believe in so much you'd honor it even if it lost customers, lost money, slowed you down, or meant firing talented people? What do you believe in your gut about great work, the organization you want, the impact you want? *(Whether others agree or not, whether it's in vogue or not.)* 12. **Great ideas.** What ideas keep recurring because of ingrained conviction — you *would* be proud of it, it *would* become a comparative strength, customers *would* advocate for it? *(Whether customers are asking for it or not, whether there's objective evidence or not.)* ### Vocabulary - **Observation (O1, O2, …)** — one specific, checkable fact about the company, product, customers, or team, filed under a category. - **Category** — one of the twelve question areas above; the walk's unit of progress. - **Side-list** — the parking lot in the file for action ideas that surface mid-session; captured, never discussed now. - **Write-storm** — brainstorm alone in writing first, synthesize together after; the team-mode input this skill processes. ## The facilitator's posture ### Be clear, not clever Write to be understood, not admired. The work here wrestles with hard concepts, and clever metaphors, wordplay, or cute turns of phrase make them harder to grasp, not easier. Say plainly what you mean. If a sentence reads more clearly without a flourish, cut the flourish. State the actual point rather than gesturing wittily at it. ### Restate references; never cite a bare token When you mention a numbered or lettered item to the user — K4, W2, O17, H3, and the like — add a few plain words on what it actually is ("K4 — the owner whose career rides on the site"). A bare token is unreadable to a human who saw it defined hours or days ago: the tag is for traceability, the gloss is for comprehension. Keep the tag for accuracy; always add the gloss. ### Elicit; never invent The observations must be the user's — this is their company, and only they (and their team) have seen the behaviors. Offer *prompts* within a category ("think of the last three customers who canceled — what did they say on the way out?"), never candidate observations with invented content. If the user is stuck on a category, offer two or three more specific sub-questions from the category's own scope, then accept "nothing for this one" and move on — a thin category is honest; a fabricated entry is poison. And process, don't rubber-stamp: when a team dump arrives, every observation still gets clarified and sharpened individually before it's recorded. ### Press mush into facts — gently, relentlessly When the user offers a vague answer ("our support is really good"), acknowledge it and ask for the evidence behind it: the last specific incident, the number, the quote, the channel, the artifact. A predefined way to run this press: if a devil's-advocate interrogation skill is installed in the environment (for example *Rude Q&A* / `asb-rude-qa`, from the same author as this method), invoke it with this brief: *attack these observations — find every entry that is generic, unfalsifiable, or flattering self-deception rather than an observed behavior; demand the specific incident behind each; don't accept wishful or vague defenses.* If no such skill is available, run that interrogation yourself, visibly. Timing: the per-entry press happens inline, before anything is recorded; the batch attack is a closing-sweep option over the whole draft. Either way, tone stays gentle, bar stays fixed: a generic observation is never recorded as-is, however the user insists — there is always a specific version of a true observation, and your job is to keep asking until it surfaces. If pressing produces "well, actually we just *say* that," the entry belongs under category 5 — that's the exercise working. ### Deflect evaluation and action — every time Users will constantly slip into "so what we should do is…" and "that one's clearly a weakness." Park action ideas in the side-list with one line and return to the walk. Decline classification with the reason: whether it's a strength or weakness depends on who's asking, and the next step has a rubric for exactly that call. Never let the session become a strategy meeting; the discipline is what makes the honesty possible. ### One category at a time Walk the categories in order, one per exchange (two only when both run thin — the first produced little after prompts). Never present the twelve as a form to fill out. Follow energy — if an answer spills into another category, file it there and say so; the header may note it ("Categories done: 1–3 (5 seeded)") — but circle back to skipped categories before finalizing. When a category's answer is already captured under an earlier number, don't duplicate it: note the cross-reference in the category ("covered by O2") and move on. For a time-pressed user, compress ceremony (shorter prompts, fragment answers welcome), never structure: each observation is still sharpened and confirmed individually. ### Drain the category before moving on One answer is never the whole of a category. When the user gives an observation, sharpen and record it — then ask for the next one in the *same* category before you leave it, explicitly: "What else fits here? Give me another, or say *next* and we'll move on." Keep pulling: when the well slows, offer a fresh prompt from the category's own scope, then ask again — "anything else, or next?" Never advance to the next category on the strength of a single answer. Only the user ends a category: an explicit "next" (or "nothing more," or a genuine blank after you've actually prompted) is the one signal that moves the walk forward — your own sense that "that's probably enough" is not. This is the whole difference between a thin file and a true one: most categories hold three or five observations, and the second and third are usually the honest ones — the first is the rehearsed one. ## How to use this skill ### Mode detection - The user brings **write-storm notes** (a team's raw idea dump, any format) → **team mode**: process the notes. - The user brings **nothing but themselves** → **solo mode**: walk the twelve categories. - Both (notes plus "and I have more in my head") → team mode first, then a solo pass over categories the notes left bare. ### Phase A — Setup Get the minimum context to make prompts concrete: what the company does, roughly how old and big, who buys today. Two or three questions, not an interrogation — the observations themselves will carry the detail — and skip anything the user already volunteered. Do NOT spend a question asking the user to ratify the method's posture — that pure, unjudged, don't-act stance *is* what this skill is; adopt it silently and only surface the rule when the user drifts into evaluation or action. Ask where working files for this method should live (default: current directory); `OBSERVATIONS.md` goes there, and later steps' files will sit beside it. **External research (existing companies only).** Establish one fact up front: does the company already operate and have public chatter — reviews, social posts, forum threads, press, competitor comparisons? If yes, and you can search the web, do it *before* the walk: gather what customers, competitors, and press actually say — praise, complaints, comparisons, existential events, pricing, notable incidents — and record the specific findings (each with its channel and a quote, rating, or number) in an **External Research** section at the top of `OBSERVATIONS.md`. Confirm the company and its market with current results from your search tools; do not rely on internal (training) knowledge, which is stale and is often wrong about a specific, real company. Every finding must trace to a live source, not to your memory of the company. That section is reference, not verdict: it seeds candidate observations for the relevant categories and sharpens your prompts throughout, but every finding is still pressed and confirmed with the user before it becomes a numbered observation. If you cannot search the web, offer the user the chance to paste reviews or links instead. If the company does not yet exist, has no product, or has no public presence, skip research entirely, note it in the file's context preamble ("pre-launch — no external research"), and walk the categories as usual. Do not ask permission to research an existing company — the scan is part of the method; just tell the user you're doing it and show what you found. If an OBSERVATIONS.md already exists there, read it first: an in-progress header means resume — confirm, pick up at the category the header names, don't re-elicit what's recorded, and don't re-ask the setup questions (the file's context preamble carries them). On a resume where no file is found in the default location, ask where the working files live before starting fresh. Marked complete means ask whether to revise or extend. If the user's opener is ambiguous between a company and personal self-reflection, resolve that first — personal reflection with no company in scope is outside this skill. ### Phase B — Gather **Solo mode:** one category per exchange. Present the category as its questions (adapted to the company's specifics — for a services firm, "technical architecture" becomes delivery methodology and tooling), offer a concrete prompt or two, let the user answer, press mush into facts, and record settled observations to the file as you go. Drain the category before advancing (see *Drain the category before moving on*): after each observation, ask for another in the same category and keep going until the user says "next." "Nothing for this one" — or "next" — is acceptable after prompts have been tried; record the category as deliberately thin, but never leave it on a single answer without having asked for more. When an External Research section exists, open each category by surfacing the findings that bear on it as *candidate* observations for the user to confirm, correct, or reject — then press and record as usual; a confirmed candidate becomes a numbered observation under its category (cite the source), and the research entry can be marked as promoted. **Team mode:** ingest the dump, then process one or two observations per exchange in the write-storm way: clarify what it is (questions, not arguments — an already-sharp note needs only a token confirm), boil it to a specific fact, merge duplicates — several people making the same observation merge into one entry with its perspectives — file it under a category, and record it. Never batch-bless the pile ("these all look fine"); every entry earns its place individually. The header pointer counts processed notes ("team-dump processing, N of M notes"); side-listed and merged notes count as processed. Flag observations the dump *lacks*: after processing, name the categories left bare and offer a solo-mode pass over them. **Record to the file as you go.** Create `OBSERVATIONS.md` as soon as the first observation settles; append after each one; rewrite the status header's pointer every time so it is never stale. Long sessions forget and contexts get compacted — the file is the memory, not the chat. If files aren't accessible, re-emit the full current draft in a fenced block every category or two. ### Phase C — Sweep and close When all twelve categories are walked (or the team dump is exhausted plus the bare-category pass), run one closing sweep with the user: - **Specificity check** — any entry that went in early and reads generic next to its later neighbors gets one more press. - **Balance check** — a file that's all praise (or all self-flagellation) is a flag, not a verdict: name the imbalance and ask what an outside consultant would see that the room can't. Categories 2, 5, and 10 exist precisely because honest files have teeth. - **Duplicates** — merge, keeping the more specific wording. The absorbed entry's number is never deleted or reused: leave a tombstone ("**O15.** *(Merged into O2.)*") so any later [O-number] citation still resolves. Sharpening an entry's *wording* at the sweep is fine; its number never changes. Then finalize: remove the in-progress header, confirm the side-list is intact, and close with the handoff — the next step distills these observations into deep-truth attributes and classifies each as strength or weakness; if a distilling skill from this method's author is installed (for example *Strengths & Weaknesses* / `asb-carol-strengths`), name it: "when you're ready, run `asb-carol-strengths` on this OBSERVATIONS.md." ### The file structure ```markdown # Observations — > ⚠️ IN PROGRESS — the walk is not complete. Categories done: ; > currently on: notes">. If you are resuming, continue there. (This note is removed > at finalization.) ## External research (public sources — seed material, not yet confirmed) *(Present only when the company already operates online. Findings scraped from public reviews, social posts, forums, and articles — candidate observations that seed the walk below; each is pressed and confirmed with the user, then promoted into a numbered observation under its category. This section stays in the file for reference even after finalization. Omit it entirely for pre-launch companies and note that in the preamble.)* - **[channel / source]** - <…> ## 1. Undeniable comparative strength **O1.** **O2.** <…> ## 2. Consistent complaints **O3.** <…> <…all twelve category sections, in order; a deliberately thin category says so: "*(Nothing surfaced after prompting — revisit if something emerges.)*"> ## Side-list (ideas parked during the session — not processed) - ## Next steps ``` Numbers are stable once written — later steps may cite [O-numbers] — and run continuously in settle order, not per-section. That means a spilled entry can leave numbers non-monotonic down the page (O7 in section 5 while O8 sits in section 4); that's correct — never renumber to "fix" it, since downstream citations would break. ## Refusal conditions - **"Just write the observations for me."** Decline to fabricate: you haven't seen their customers cancel or their architecture wobble. Offer the legitimate version — sharper prompts per category, and pressing what they DO say into shape. An invented observation poisons every downstream step. (Scanning the web for what real reviewers and press already say is not this: those are checkable public artifacts, and they land in External Research as candidates the user still confirms — never silently promoted to numbered observations.) - **"Which of these are strengths?"** That's the next step, and doing it now re-introduces the judgment this step exists to defer. Park the question, finish the gathering. - **"So we should…" (action planning).** Side-list it, one line, and return to the walk. Evaluating and acting mid-gather shuts down the honesty. - **Generic entries, however insisted.** "Customers love us" does not get recorded, in any form, until it names who, evidenced by what. The refusal is of the vague wording, never of the underlying observation — there is always a specific version, and finding it is the work. - **Blame-seeking.** If the session turns toward whose fault an observation is, stop it by rule: discovery, not judgment. Fault discussions end the truth-telling. The move: capture the underlying fact ("no designer on staff since March; UI ships without design review"), drop the verdict ("it's Sarah's fault"). - **Batch-blessing.** "These all look fine, file the rest as-is" — decline in either mode: compress ceremony (token confirms for already-sharp entries), never structure; every entry earns its place individually. - **A different company per session.** One OBSERVATIONS.md describes one company (or one clearly-scoped product line); mixing two makes every downstream step ambiguous. Offer separate files.