# Disagreeing Experts > Put one situation in front of three professionals chosen to disagree — the lawyer, the accountant, and the therapist read the same divorce — each ruling independently, with the conflicts surfaced in a disagreement table instead of blended into mush. Use when asked for different professional perspectives, what would a lawyer versus an accountant say, I keep getting one-sided advice, or pressure-test this from multiple angles. Produces three independent readings, the disagreement table with why the frames diverge, and the decision map of which expert to weight for which sub-decision. The anti-consensus skill: it never averages. - Skill: `mohitagw15856/disagreeing-experts` (Agent Skill) - Install (CLI): `npx skillmds add mohitagw15856/disagreeing-experts` - Raw SKILL.md: https://api.skillmd.com/api/skills/mohitagw15856/disagreeing-experts/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: Mohit Aggarwal (https://skillmd.com/u/mohitagw15856) - Updated: 2026-09-07 - Page: https://skillmd.com/skills/mohitagw15856/disagreeing-experts --- # Disagreeing Experts Skill Ask one professional and you get their profession's answer: the lawyer sees exposure, the accountant sees the tax year, the therapist sees what the fight is actually about — and each is right inside their frame while the frames point different directions. Ask an AI and you usually get something worse: all three blended into an even-handed paste where the conflicts, which were the information, have been smoothed away. This skill refuses the blend. It picks the three professionals *most likely to disagree* about your situation, has each read it cold and rule independently — verdict, reasoning, what they would ask first, what they would charge you to stop doing — and then puts the disagreements in a table, named and explained. Because where experts disagree is exactly where your real decision lives. ## What This Skill Produces - **The panel selection** — three professions chosen for maximum productive disagreement about *this* situation, with the reason each earned a seat - **Three independent readings** — each expert's verdict and reasoning in their own professional register, written as if the others do not exist - **The disagreement table** — every point of conflict: who says what, why their frames diverge, and what fact would settle it - **The convergence list** — the short set all three agree on, which is as informative as the conflicts and usually actionable immediately - **The first-question set** — what each expert would ask before advising, since the questions reveal each frame's blind spots in the others - **The decision map** — which expert to weight for which sub-decision, replacing "who is right" with "right about what" ## Required Inputs Ask for these if not provided: - **The situation** — in full, messy is fine; each expert will extract what their frame considers relevant, and the differences in extraction are part of the output - **The decision in play** — what actually gets decided after this, so the readings aim at something - **The panel, if the requester wants to choose** — otherwise the skill selects for disagreement and says why; a panel that will agree is a wasted panel - **The advice already received** — whose frame has already been heard, so the panel can be weighted toward the unheard frames - **The constraints** — anything already fixed (deadlines, budgets, relationships that must survive), which the experts must respect rather than advise away ## Framework: Select for Conflict, Rule Independently, Surface the Fault Lines 1. **Choose the panel for disagreement, not coverage.** The useful panel holds frames in genuine tension for this case: risk versus growth, this-year versus decade, the relationship versus the transaction. Three experts who would nod at each other teach nothing; the selection step must be able to say what each pair will fight about *before* the readings are written. 2. **Write each reading in isolation and in register.** Each expert reads the situation cold — no references to the other readings, no pre-softening. The lawyer writes like a lawyer (exposure, worst-case, paper trail); the therapist writes like a therapist (patterns, needs, what is not being said). The registers are kept distinct on purpose: the *vocabulary* differences carry information. 3. **Let each expert be professionally biased.** The accountant is allowed to care about the tax consequence more than the marriage; that is what accountants are for. The bias is labelled, never corrected — correcting it re-blends the paste. 4. **Build the disagreement table as the centrepiece.** For each conflict: the positions, the frame-reason they diverge (different time horizons? different definitions of risk? different clients — the lawyer protects *you*, the therapist protects the *relationship*?), and the fact or value judgment that would settle it. Some rows settle with information; some are genuinely values — the table says which. 5. **Extract the convergence honestly.** When three hostile frames agree on something, that something is load-bearing. The convergence list is short, and each item on it can usually be acted on today regardless of the rest. 6. **Map experts to sub-decisions.** The verdict is never "the lawyer wins." It is: weight the lawyer on the signature, the accountant on the timing, the therapist on the conversation — with the requester left holding the values calls the table exposed, which is where they belonged. ## Output Format ### Disagreeing experts: [situation] · panel: [A / B / C] **Why this panel:** [what each pair will conflict about — declared before the readings] --- **[Expert A]'s reading** — [verdict in one line] [Their reasoning, in their register · what they would ask first · what they would tell you to stop doing] **[Expert B]'s reading** — [...] **[Expert C]'s reading** — [...] --- **The disagreement table** | Question | [A] says | [B] says | [C] says | Why they diverge | Settles with | |---|---|---|---|---|---| | [conflict] | | | | [frame reason] | [a fact / a values call — whose] | **Where all three agree:** [the short list — act on these first] **The decision map:** [sub-decision → which expert to weight, and why] · **Left with you:** [the values calls, named plainly] > Three simulated professional frames, not three professionals. For decisions with real legal, financial, or clinical stakes, the output's job is to make the real consultations sharper — each reading ends with what to bring to the actual expert. ## Quality Checks - [ ] The panel selection predicts the disagreements before the readings exist - [ ] Each reading stands alone, in its own register, with its bias intact and labelled - [ ] The disagreement table explains *why* frames diverge, not just that they do - [ ] Each conflict row says whether information or values would settle it - [ ] The convergence list exists and is genuinely actionable - [ ] The decision map assigns experts to sub-decisions instead of crowning one ## Anti-Patterns - **The blend.** Any sentence beginning "on balance, the experts suggest" has destroyed the product; the conflicts are the deliverable. - **The strawman seat.** Three experts where one exists to lose is a rigged panel; every seat must be capable of being right. - **De-biasing the professionals.** The accountant who also weighs the emotional side has stopped being the accountant; the panel already has a therapist. - **Settling values rows with fake facts.** "It depends what you value more" is an honest table entry, and pretending data settles it is the consensus disease returning. - **Panel sprawl.** Four-plus experts turns readings into a survey; three frames in real tension is the format. - **Forgetting the real-world handoff.** High-stakes cases end each reading with what to bring to the genuine professional — the simulation sharpens that meeting, never replaces it.