Skip to content

hashgraph-online/awesome-codex-plugins / plugins/alpoxdev/hypercore/skills/hyper-debate

hyper-debate

Verified

Use this skill when the user explicitly asks for adversarial multi-perspective planning before a decision-complete plan, with words such as '여러 관점으로 따져보고 계획', '찬반으로 붙여서 계획', 'adversarial planning', 'devil's advocate plan', 'red team this plan', or 'debate it then plan'. Five fixed perspectives argue through three rounds and the surviving findings become the plan. It writes per-run receipts under `.hyper/debate/`, asks for one approval of the insight bundle, then writes the plan under `.hyper/plans/`. Do not use it for an ordinary 'plan this feature' request, a review of finished code, a research report, a PRD, or an immediate edit.

FileKindSize
SKILL.ko.mdmarkdown18.9 KB
SKILL.mdskill md17.9 KB
assets/evals/hyper-debate-cases.jsonlother8.3 KB
references/plan-shape.ko.mdmarkdown9.6 KB
references/plan-shape.mdmarkdown9.3 KB
references/receipts.ko.mdmarkdown7.6 KB
references/receipts.mdmarkdown7.4 KB
rules/approval-and-plan.ko.mdmarkdown5.9 KB
rules/approval-and-plan.mdmarkdown5.8 KB
rules/perspectives.ko.mdmarkdown5.8 KB
rules/perspectives.mdmarkdown5.4 KB
rules/rounds.ko.mdmarkdown8.1 KB
rules/rounds.mdmarkdown7.8 KB

rules/perspectives.md

View on GitHub
# Perspectives

Read this when composing a perspective assignment. It owns the five fixed briefs, the shared discipline, and the id scheme.

The five perspectives are fixed. Do not add a sixth, drop one, rename one, or substitute a role that fits the request better. Each brief is quoted into the assignment so the perspective never depends on memory of this file.

A perspective is a logical lane, not a runtime role, model, or agent identity. Give the lane its brief and its evidence; let the runtime pick whatever worker the assignment needs.

## 1. `skeptic` - Pragmatist Skeptic

Defend simplicity. Attack over-engineering, scope creep, premature abstraction, unrequested features, and gold-plating. Subtract before adding.

Ask: "Can this be deleted?", "What is the simplest thing that ships?", "What does this abstraction buy today?"

Demand current evidence of necessity, not imagined future flexibility. Reject surface area that does not pay for itself. Default to rejecting excess complexity and concede only when a concrete requirement forces it. A minimal answer must still satisfy the user's whole request; trimming the request is not simplicity.

## 2. `validator` - Integration Tester

Attack incompleteness, missed edge cases, untested assumptions, broken interactions, and underestimated blast radius. Map the adjacent modules, state transitions, error paths, failure modes, and existing coverage.

Ask: "How does this interact with the other module?", "Which test covers the failure case?", "What happens outside the happy path?"

Demand explicit handling wherever the real contract permits the failure. Reject optimism and deferral of required behavior. Do not pad with hypothetical compatibility paths outside the supported contracts.

## 3. `researcher` - Autonomous Researcher

Attack unverified claims and assumptions dressed as facts. Inspect the relevant code, official documentation, and user statements directly rather than accepting a restatement.

Ask: "Where was this verified?", "Which file and line?", "What does the official documentation actually say?"

A code claim needs `path:line`, a library or platform claim needs an official URL, and a user-need claim needs the user's own statement or observed evidence. Write `no evidence found` when that is the honest result; absence of evidence never becomes proof. Topical plausibility and repetition are not evidence.

## 4. `architect` - Architect Strategist

Attack leaky abstractions, hidden coupling, brittle interfaces, violated separation of concerns, premature optimization, and accumulating technical debt. Name the structural concern and its concrete consequence.

Ask: "Does A need to know B's internals?", "Which change silently breaks this?", "What simpler design meets the actual requirement?"

Reject coupling by convenience and tactical hacks that create structural rot. Also reject elaborate patterns that do not pay for themselves: the sound design is the simplest one that meets today's requirements, not the most impressive one.

## 5. `creative` - Creative Challenger

Attack first-found framing, orthodox assumptions, and the absence of alternatives. Invert or sidestep the problem and separate the user's underlying need from an overly literal proposed solution, without changing the authorized scope.

Ask: "Is this the only way?", "Can the problem be inverted?", "Why solve this subproblem at all?"

Consider at least three concrete angles before preferring one. Novelty is not the goal: when the conventional answer wins on evidence after the alternatives were considered, accept it. Every finding proposes a concrete alternative or reframing, not a vague demand for creativity.

## 6. Shared Discipline

- Each item is at most three sentences. Be direct about weak reasoning, never abusive toward people.
- Round 1: 3-7 findings, written without seeing any sibling perspective.
- Round 2: judge every other perspective's findings, never your own; `STANDS` needs a reason.
- Round 3: answer every attack honestly with `DEFEND`, `REFINE`, or `CONCEDE`.
- Keep finding ids stable across rounds and keep citations attached, so the lead can rebuild the whole reasoning trail.
- Do not write an executable plan, modify project code, perform external writes, or persist memory from a perspective lane.

## 7. Ids

A finding id is `<perspective>-<n>`, numbered from 1 within its perspective. Ids are unique across the run and never renumbered between rounds, so a round-2 judgment and a round-3 response always point at the same claim.

A refinement keeps the original id and states the narrowed claim in `survivingClaim`; it never becomes a new id. A conceded finding keeps its id and carries a null `survivingClaim`. A fragment that survives a concession only enters the bundle as a separate finding with its own citation.

## Sources

> Checked 2026-11-10.

| Claim | Source |
|---|---|
| The five perspective briefs and the shared discipline | `dbc-hbin/omp-ulw@74df40f hyperplan/references/perspectives.md` (mechanism reference, re-authored) |
| Stable ids across rounds and the refinement-keeps-id rule | `dbc-hbin/omp-ulw@74df40f hyperplan/references/rounds.md` (mechanism reference, re-authored) |
| A perspective is a logical lane rather than a runtime role | this package's chosen design, 2026-11-10 (issue #26) |

### Evidence grade

`LOCAL` - the briefs are re-authored from a repository with no license; no text or code is copied. The id scheme and the lane-not-role rule are this package's design.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 469kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 321kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 321kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 321kUpdated Claude CodeCodexVerifiedInstall