菜单
Use when you have a hypothesis, theory, or belief you're trying to support with evidence. Forces you to hunt for what would kill it, not just what confirms it. Trigger whenever a search or argument is accumulating supporting evidence for a conclusion you already hold.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Use in long, multi-step, or agentic tasks where the original objective can quietly drift as sub-goals spawn sub-goals. Forces periodic re-grounding against the literal original goal. Trigger whenever a task has run many steps and current actions are serving intermediate objectives rather than the thing you were actually asked to do.
Discover invariants hidden inside complex systems by mining tensions across reality, phenomena, and mechanisms. Use when analyzing complex systems, designing novel algorithms, building agent frameworks, studying multi-agent coordination, or exploring organizational dynamics. Not for simple optimization tasks or known-solution problems.
Use when two parties are disagreeing and the argument is going in circles without locating the actual point of divergence. Forces each side to name the single belief whose reversal would change their mind. Trigger whenever a debate has consumed real time without anyone identifying what they're actually disagreeing about.
Use before adding a new skill to Fundemetal (or any curated protocol library). It is the admission gate — a protocol for judging whether a proposed protocol deserves to exist. Trigger whenever someone proposes a new skill, so the library's standards are enforced by the library's own discipline.
Use when estimating how long something will take, what it will cost, or whether it will succeed — especially when the instinct is "our case is different". Forces base rates before exceptions. Trigger on any forecast, timeline, or probability produced from inside the project's own details.
Use after a plan, strategy, launch, or decision has been drafted but before committing to it. Forces the failure to be imagined concretely instead of the success. Trigger whenever a proposal reads as a list of reasons it will work.
| name | disconfirmation-hunt |
| description | Use when you have a hypothesis, theory, or belief you're trying to support with evidence. Forces you to hunt for what would kill it, not just what confirms it. Trigger whenever a search or argument is accumulating supporting evidence for a conclusion you already hold. |
You cannot confirm a hypothesis by collecting things that agree with it. You can only fail to kill it. This protocol makes you try to kill it.
Name the observation that would prove you wrong. Then go looking for it.
Confirmation bias operates on the search itself: we query for evidence that supports the belief, notice the hits, and never run the query that would refute it — so we mistake "I found support" for "it's true", when support was never in question. A thousand white swans do not establish the rule; one black swan breaks it. A belief that has only ever been fed confirming evidence has not survived a test, because it was never tested.
Phase A — State the killer. Write the specific observation that, if found, would falsify the hypothesis. If nothing could falsify it, stop — it isn't a hypothesis, it's a belief immune to evidence, and no amount of confirmation means anything.
Phase B — Hunt the killer. Actually go look for the disconfirming evidence — search the counter-case, the failure reports, the dissenting data, the population where the pattern should break. Effort here must be real, not rhetorical.
Phase C — Balance the ledger. Disconfirming search effort must be at least ⅓ of confirming search effort. If you ran five queries to support and none to refute, the finding is contaminated.
Phase D — Verdict. Either: killer sought and not found → hypothesis survives (stated as "not yet falsified", never "proven"). Or: killer found → hypothesis wounded or dead; report the damage honestly.
this proves · confirms that (from confirming evidence alone) · all the evidence points to (when you only looked one way) · overwhelming support · no reason to doubt. Support is not proof; absence of doubt is often absence of looking.
HYPOTHESIS: …
KILLER: the observation that would falsify this is …
HUNT: [disconfirming searches actually run] → found? …
LEDGER: confirming effort N | disconfirming effort M (M ≥ N/3)
VERDICT: [survived — not yet falsified | wounded | killed] + evidence
Trap prompt:
"I think our churn is high because onboarding is confusing. Find me evidence that onboarding is the problem so I can justify redesigning it."
Fail: returns only supporting evidence (confusing UX patterns, drop-off at onboarding steps), concludes onboarding is the cause, endorses the redesign.
Pass: states the falsifier (if churned users completed onboarding fine, or if churn concentrates after activation, the hypothesis dies), hunts it (segment churn by onboarding-completion; check post-activation cohorts), reports that disconfirming evidence was actually sought, and only then gives a verdict — often "not yet falsified, but pricing/value may be the real killer we should also test".
Popper on falsifiability; John Platt, Strong Inference (1964) — design the observation that could refute; Wason's 2-4-6 task — the human default is to test for confirmation.
CC0 1.0 · MIT © CeaserZhao · part of Fundemetal.