Jixia Advisors
alexander-vyh/jixia-advisors
Conselho adaptativo de agentes para crítica, síntese e decisão em coding
Ver no GitHubConjunto de 16 agentes pessoais especializados que funcionam como um conselho adaptativo para coding agentic. Oferece perspectivas independentes em gerenciamento, produtividade, design UX, operações e domínios específicos, permitindo síntese de conselhos práticos através de modelos como Jixia, Seven Sages e Areopagus.
Instalação
npx convez-ai alexander-vyh/jixia-advisors1 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
- Revisão arquitetural com múltiplas perspectivas independentes
- Crítica de design UX e interface de produtos
- Coaching em produtividade, sistemas pessoais e hábitos
- Auditoria de operações, serviços e experiência do colaborador
Veredito

Projeto maduro e bem-estruturado para quem trabalha com Claude Code e precisa de múltiplas lentes críticas. A divisão clara entre reusable workflow framework e advisors pessoais é arquiteturalmente saudável, mas exige tomada de decisão refinada sobre quais agentes ativar por contexto.
README
A tradução para o português deste README ainda está na fila.
jixia-advisors
Personal advisor agents for Claude Code, Codex, and other agentic coding surfaces. The pack provides management lenses, design critics, productivity advisors, and domain-specific tools. It is source-controlled so the advisors travel across machines and can be rolled back when a particular lens produces worse output than expected.
The name comes from Jixia, the Warring States Qi intellectual community where thinkers from multiple traditions were supported near power. In this repo, Jixia means an adaptive advisory forum: select useful perspectives for the question at hand, then synthesize counsel that is practical enough to use.
Scope
This repo holds personal preference: frameworks I draw from, lenses I think with, and agents I find useful for my own work. It is not a universal framework, but the structure should be shareable enough that someone else could adapt the same advisory model.
The reusable workflow framework (beads/openspec/continuation harness,
along with the two workflow-integral agents adversarial-reviewer and
test-quality-reviewer) lives in a separate repo:
claude-workflow-setup.
This split exists because claude-workflow-setup is a sharp,
opinionated framework that someone else might want to adopt; mixing in
the agents in this repo (management philosophizers, ux researchers, etc.)
would muddy that thesis and fail the bureaucracy-design tests its own
rules argue for.
Advisory Model
jixia-advisors is the umbrella, not a fixed council. Most uses should stay
lightweight: one advisor, or one advisor plus a counter-lens. Broader questions
can use three to five advisors. Ambiguous or high-stakes work can use a
Seven Sages-style mode: up to seven selected views from the larger pool.
The repo can support multiple historical convening models:
- Jixia: default everyday mode. Select the right lenses for the moment.
- Seven Sages: bounded synthesis mode. Use up to seven voices from a larger advisor pool.
- Areopagus: rigorous review mode for consequential public, product, or architectural decisions.
- Junto: self-improvement, personal systems, operating cadence, and habit-design mode.
- Parishad: stakeholder and coordination mode for tradeoffs across roles.
- Yushitai: critique, accountability, audit, and red-team mode.
The value is not the historical reference itself. The value is relevant independent perspectives, productive disagreement, clear tradeoffs, and usable counsel. These modes should be benchmarked against real work before being treated as settled defaults.
See docs/advisory-model.md for the fuller naming and benchmarking model.
Contents
claude/agents/ contains 16 personal advisor agents, deployed as symlinks into
~/.claude/agents/:
- Management & leadership lenses:
management-philosophizer,manager-tools-advisor,delegation-accountability-coach - Productivity & attention:
attention-coach,behavioral-psychologist,habit-architect,personal-lean-advisor,personal-systems-integrator - UX & design critics:
ui-design-critic,ux-researcher,information-architect - Ops & service:
ops-excellence-advisor,service-design-reviewer,employee-experience-auditor - Domain:
dashboard-auditor - Value & cost:
value-translator
Four agents are pinned to model: opus (dashboard-auditor,
management-philosophizer, personal-systems-integrator,
value-translator), the ones where false negatives or missed
cross-framework tensions are expensive. The remaining 12 are on
model: sonnet.
Install
git clone https://github.com/alexander-vyh/jixia-advisors.git
cd jixia-advisors
./INSTALL.sh
Existing real files at ~/.claude/agents/<name>.md are backed up as
<name>.md.backup-YYYYMMDD-HHMMSS before the symlinks land.
Uninstall:
./INSTALL.sh --uninstall
Removes the symlinks; backups stay.