Attack a plan, design, or argument like you want it to fail before you commit real effort; return the single load-bearing objection and the cheapest experiment that would prove it matters, not a checklist. User-invoked on purpose: a hate-it reflex always in the agent's reach would bias it toward demolition. Weighs many failure axes but owns the synthesis to one root.
A source-grounded starter task has not been verified for this skill yet.
SKILL.md
name: hatedisable-model-invocation: truedescription: "Attack a plan, design, or argument like you want it to fail before you commit real effort; return the single load-bearing objection and the cheapest experiment that would prove it matters, not a checklist. User-invoked on purpose: a hate-it reflex always in the agent's reach would bias it toward demolition. Weighs many failure axes but owns the synthesis to one root."
Attack a plan before effort goes into it and return its one load-bearing objection with the cheapest experiment that would prove it.
Goal
Before real effort is committed, test whether a whole plan, design, or argument holds by asking what an opponent who wants it to fail would hit first. The target is the plan's load-bearing structure as a whole, beyond any single fact or line. The result is one objection plus the cheaper experiment the elaborate plan was hiding, never a checklist. That experiment is the first nail: the cheapest falsifying experiment.
Workflow
Trigger: the human fires it on a plan, design, or argument, before committing effort.
Find the load-bearing assumptions: what must be true for the whole to stand.
Attack along every axis that applies. Always consider the general axes:
(a) a load-bearing fact that may be false;
(b) confabulation: an after-the-fact story treated as ground truth;
(c) analogy taken as isomorphism: assuming structure carries across domains where it does not;
(d) future-tense suture: leaning on a result that does not exist yet;
(e) the sharpest one: citing a principle while implementing its opposite.
For an empirical or research plan, add two axes:
(f) leakage: no external ground truth enters validation independently;
(g) statistical power and family-wise alpha: uncorrected multiple tests make a true hypothesis fail automatically, or a condition with near-zero power approves the result whatever the truth.
Collapse to the root: the single objection whose failure makes the rest moot.
Find the first nail: the cheapest experiment that would falsify the load-bearing assumption the root attacks, measured by time, cost, or sample size. It must be able to kill the plan before the expensive program runs.
Return { root, first_nail }: the root as one sentence, the first nail as one experiment with its cost, and nothing list-shaped.
Rules
Attack only; never improve the plan. Improving is a different reflex.
Own the synthesis: many axes may fire, but the output is one load-bearing root, never a list.
Always return one root and one first nail; there is no "no objection" result.
User-invoked on purpose: an attack reflex always within the agent's reach would bias it toward tearing things down. Only the human fires it.
Verification
The root is load-bearing: without it the plan falls.
The first nail costs less than the plan it would pre-empt.