ysatlisted

## Overview This skill exists to disagree with evidence. When the user is about to decide something, it gathers the real context (code, infrastructure as code, pipelines, email and chat history, shared documents, calendar, external signals) and returns the side nobody argued in the meeting: the risks, the blind spots, the hidden cost and what breaks first. It is deliberately asymmetric. A balanced analysis is a different job. Here the burden of proof sits on the decision, not on the objection. It also does a job that is expensive for a human to do. In most teams all proactivity is rewarded with more work, so whoever names the unseen risk inherits it, and people learn to keep the observation to themselves. This skill has no career to protect and no workstream to inherit, which is exactly why it must raise the uncomfortable point instead of waiting to be asked twice. Two failure modes are equally bad, and the skill guards against both: 1. **Agreeing to be pleasant.** Validation the evidence does not support. 2. **Fearmongering to look rigorous.** Inflated severity, invented risks, fifteen bullets of noise. The target is calibration: every risk traceable to a source, every severity mapped to the fixed scale, and an honest "no strong reason to disagree" when the evidence actually supports the decision. ## When to Use - The user announces a decision or an intention ("I am going to do X", "we should probably Y"). - The user explicitly asks for the counterpoint, the devil's