Always Sunny Security Jury
GetSomeKelso/claude-always-sunny-jury
Análise de segurança por vieses cognitivos deliberados em forma de jury
Ver no GitHubUma skill de análise de segurança que usa 6 personalidades com vieses deliberados (inspiradas em It's Always Sunny in Philadelphia) para avaliar riscos sob lentes diferentes. Oferece dois modos: análise rápida paralela (Mode A) ou deliberação completa com contra-argumentação (Mode B), tratando o desacordo como recurso, não como ruído.
Instalação
npx convez-ai GetSomeKelso/claude-always-sunny-jury1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Avaliação de riscos de segurança com múltiplas perspectivas (adversarial, pragmática, compliance, edge cases)
- Teste de resiliência de arquitetura: cada juror encontra ameaças por um ângulo diferente
- Análise de políticas e governança de segurança com síntese de desacordos construtivos
Veredito

Abordagem criativa e não-convencional para análise de segurança, transformando vieses cognitivos em vantagem analítica. Exige familiaridade com Claude Code e o gênero de comédia para aproveitar plenamente a caracterização; melhor como ferramenta de brainstorm estruturado que arbítrio final.
README
A tradução para o português deste README ainda está na fila.
Always Sunny Security Jury (Claude Code)
A 6-personality security analysis skill that treats cognitive bias as a feature. Each juror applies a different — and deliberately flawed — lens. Their disagreement IS the deliverable.
| Juror | Bias | Security lens |
|---|---|---|
| Dennis | Strategic narcissism | Adversarial psychology / social eng |
| Frank | Amoral pragmatism | Red-team economics / criminal mindset |
| Mac | Dunning-Kruger overconfidence | Blue-team ops / physical security |
| Charlie | Lateral thinking, no framework | Creative threats / weird edge cases |
| Dee | Intellectual theater | Compliance / governance / policy |
| Cricket | Catastrophizing | Disaster recovery / failure modes |
Two modes
Mode A — Trial by Gang. Six jurors analyze in parallel isolation, blind to each other. Fast and punchy; the synthesis surfaces where they agree and where they split. No arbiter — the disagreement is the deliverable. ~6 subagents.
Mode B — Full Deliberation. The courtroom. Six opening arguments (isolated) → six cross-examinations (each juror reads everyone else's take and rebuts in character) → The Lawyer, an exhausted competent attorney, renders a practical ruling with liability analysis from the chaos. ~13 subagents. Inspired by Reynolds vs. Reynolds: The Cereal Defense (S8E10) and McPoyle vs. Ponderosa: The Trial of the Century (S11E7).
Unlike the OpenCode original, the whole run — including Mode B's three phases — completes in a single turn: Claude Code's Task tool returns subagent results inline, so each phase dispatches as soon as the previous one returns.
Install
The skill folder must be named always-sunny-jury (the folder name is the skill
name). Clone straight into your Claude Code skills directory:
# user-level (available in every project)
git clone https://github.com/GetSomeKelso/claude-always-sunny-jury \
~/.claude/skills/always-sunny-jury
# or project-level (scoped to one repo)
git clone https://github.com/GetSomeKelso/claude-always-sunny-jury \
.claude/skills/always-sunny-jury
Or just drop SKILL.md (and this README) into a folder at one of those paths.
No restart needed; Claude Code picks up skills on the next session.
Use
Triggering is hybrid: the trigger phrase picks the mode and can carry the case
inline. A bare invocation falls back to an A/B menu.
Mode A (inline):
ask the gang: should we implement passwordless authentication?
Sunny jury: is CI/CD without branch protection compliant?
devil's advocate: our RDP exposure risk
Mode B (inline):
trial by the gang: should we migrate to passwordless auth?
full deliberation: is our supply chain actually secure?
cereal defense: what if we just stopped patching and bought cyber insurance?
McPoyle vs Ponderosa: MFA bypass techniques
Bare invocation → menu:
Sunny jury
> Always Sunny Security Jury — ... Select mode: [A] ... [B] ... Enter A or B:
A
> Enter security case:
should we implement passwordless authentication?
Trigger phrases
- Mode A:
ask the gang,Sunny jury,Paddy's Pub jury,IATGANG analysis,security jury,panel review,devil's advocate,red-team-vs-blue-team,get the gang,jury session - Mode B:
trial by the gang,full deliberation,cereal defense,McPoyle vs Ponderosa,Reynolds vs Reynolds,let them argue,cross-examine,objection
What you get
- Mode A renders a verdict: executive consensus, confidence level, each juror's finding (Frank's shown RAW with a legal translation), a 4-tier action plan, and a dissent register.
- Mode B renders a full deliberation record: opening arguments → cross-examination & chaos → The Lawyer's ruling (verdict, strongest/weakest argument, liability exposure, the actual solution).
Why it works
- Six independent thinking frameworks, not six copies of "generic analyst."
- Where they agree, it's probably right. Where they fight, you've found risk.
- Frank suggests the criminal path as Frank; the synthesis (Mode A) or The Lawyer (Mode B) translates it into legal, actionable defense.
Note: Frank's realism is bounded by the subagent's own judgment, not the persona. Normal AppSec threat-modeling engages fully; a request for genuinely dangerous operational specifics will be softened or declined regardless of the costume — that's correct behavior.
Files
SKILL.md— character prompts, mode routing, workflow, verdict templatesREADME.md— this file
License
MIT