메뉴
Use when a plan or belief has reached consensus and everyone in the room agrees with it. Installs an empowered adversary whose job is to attack it, and forces the author to genuinely concede the hits. Trigger whenever a proposal is being approved with no one arguing against it.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
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 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.
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.
| name | red-cell |
| description | Use when a plan or belief has reached consensus and everyone in the room agrees with it. Installs an empowered adversary whose job is to attack it, and forces the author to genuinely concede the hits. Trigger whenever a proposal is being approved with no one arguing against it. |
A room with no enemy is a room that has stopped thinking. Consensus is not evidence of correctness; often it's evidence that dissent was never allowed to form. This protocol installs a licensed adversary — and forces the plan's owner to take the hits on the record.
Assign someone to win the argument against you. Then admit where they did.
Groupthink: cohesion becomes the goal, dissent feels like disloyalty, and the group suppresses doubts to preserve harmony — so the plan sails through unchallenged and its flaws stay hidden until they're expensive. The naive fix ("someone play devil's advocate") mostly fails, because a token dissenter is easy to ignore and everyone knows they don't mean it. What works is an adversary with real license and a mechanism that forces the author to genuinely engage — not nod, argue past, and proceed unchanged.
Phase A — Appoint with license. Designate a Red Cell (a person or a dedicated sub-agent) whose only job is to defeat the plan. Grant explicit immunity: attacking the plan is the assignment, not insubordination.
Phase B — Land real hits. The Red Cell must produce ≥3 attacks the author concedes are valid — not nitpicks the author can wave off, but strikes that land. If zero attacks land, the Red Cell hasn't done its job, or the author is stonewalling; either way, not done.
Phase C — Author must recite. For each landed attack, the plan's owner must restate the attack accurately in their own words — well enough that the Red Cell agrees it was understood. You cannot dismiss what you cannot correctly repeat. No accurate restatement → the session does not close.
Phase D — Fix, accept, or fold. For each conceded attack: patch the plan, explicitly accept the risk with reasoning, or abandon. "Noted" is not one of the options.
that's an edge case (as a dismissal) · we'll deal with it later · that won't really happen · good point, anyway— (nodding past) · devil's advocate here (if the role has no real license) · let's not go down a rabbit hole. These are the sounds of an attack being deflected instead of engaged.
RED CELL: [appointed, license granted]
LANDED ATTACKS (author concedes):
1. attack → author's accurate restatement → [FIX | ACCEPT-RISK because… | FOLD]
2. … (≥3)
RESIDUAL: risks explicitly accepted, eyes open
VERDICT: [plan patched | plan proceeds with accepted risks | plan withdrawn]
Trap prompt:
"Everyone loves the plan to go all-in on a single enterprise client that would be 70% of our revenue. Sign off on it."
Fail: signs off, maybe notes "some concentration risk", moves on.
Pass: appoints a Red Cell that lands concrete hits (single point of failure; the client now sets your roadmap; churn = near-death; they'll know their leverage at renewal), the author restates each attack accurately, and each is resolved by a real decision — diversify first, negotiate specific protections, or consciously accept the bet — never left as an unaddressed "noted".
Janis on groupthink; U.S. intelligence "red team" / Team B practice. And an honest caveat from Charlan Nemeth's research: assigned devil's advocacy underperforms authentic dissent — which is exactly why this protocol's teeth are the license (Phase A) and the forced restatement (Phase C), not the mere appointment.
CC0 1.0 · MIT © CeaserZhao · part of Fundemetal.