convez.ai
AgenteAutomaçãoAvançado
neolin0629

AI Instructions Workspace

neolin0629/ai-instructions

Workspace centralizado para instruções e agentes IA multilíngues

Ver no GitHub
InicialPluginScore · de 0 a 100

Repositório centralizado para gerenciar e padronizar instruções de agentes de IA e prompts de sistema para ferramentas como Claude Code, Codex CLI e Gemini. Organiza hierarquicamente instruções multilíngues (EN/ZH) com roteamento de agentes especializados (code_dev, code_review, writer, architect, product-manager) e configurações de runtime.

Instalação

$npx convez-ai neolin0629/ai-instructions

1 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

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 1 semana
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Padronizar comportamento de agentes de IA em equipes
  • Gerenciar instruções específicas para diferentes personas (dev, reviewer, writer, architect)
  • Manter configurações consistentes entre Claude, Codex e Gemini
  • Implementar roteamento inteligente de tarefas para agentes especializados

Veredito

Projeto bem estruturado para times que precisam orquestrar múltiplos agentes de IA com instruções padronizadas e roteamento inteligente. Suporta multilíngue e múltiplas plataformas (Claude, Codex, Gemini), ideal para organizações escalando uso de IA.

README

A tradução para o português deste README ainda está na fila.

AI Instructions Workspace

This repository serves as a centralized workspace for managing, standardizing, and deploying AI agent instructions and system prompts for different CLI tools (like Codex CLI and Claude Code).

Directory Structure

The repository is organized by tool and language. Language-specific directories use the same filenames to maintain consistent cross-referencing:

  • codex/: English instructions and configurations for Codex CLI.
  • codex_zh/: Chinese instructions and configurations for Codex CLI.
  • claude/: English instructions for Claude Code.
  • claude_zh/: Chinese instructions for Claude Code.
  • gemini/: English instructions for Antigravity (Gemini).
  • gemini_zh/: Chinese instructions for Antigravity (Gemini).

Core Files & Architecture

The instruction architecture follows a hierarchical design where the entry file handles both global rules and the routing of specialized sub-agents.

1. AGENTS.md / CLAUDE.md / GEMINI.md (Universal Entry & Routing)

This is the global entry point and primary rulebook. It handles:

  • Core Behavior: Communication language, fundamental principles (e.g., "Think before coding", "Simplicity first").
  • Agent Routing & Registry: Defines how the system identifies and activates specialized agents (code_dev, code_review, writer) based on task types and keywords.
  • Global Baselines: Universal coding standards (e.g., Python 3.10+), data processing rules, and Git safety protocols.
  • Output Contract: The expected format and verification requirements for final responses.

2. config.toml (Runtime Configuration)

For Codex, this file is strictly for system and performance settings:

  • Concurrency: max_threads for parallel task execution.
  • Depth: max_depth for agent recursion limits.
  • Note: Agent registration and keyword routing have been moved to AGENTS.md for better central management.

3. agents/*.toml or agents/*.md (Specialized Personas)

These files define the fine-grained behavior, checklists, and technical priorities for specific roles.

  • product-manager (Product Management):
    • Requirement convergence, PRD authoring, and market/competitive research. Only writes PRD documents.
  • architect (System Architecture):
    • System design, technology selection, data modeling, and task breakdown. Only writes design documents.
  • code_dev / code-dev (Code Development):
    • Implementation, debugging, performance optimization, and quant backtesting.
    • Correctness > Performance > Simplicity.
  • code_review / code-review (Code Review):
    • Independent auditing and risk assessment. Read-only; emphasizes edge cases and correctness.
  • writer (Content Writing):
    • General document writing (articles, reports, explainers, docs, notes). Emphasizes typography and factual integrity.

How It Works (Read Order)

Codex CLI

  1. AGENTS.md: Establishes ground rules and performs Agent Routing.
  2. config.toml: Applies runtime limits (threads/depth).
  3. agents/<agent>.toml: Loads the selected agent's specific instructions.

Claude Code

  1. CLAUDE.md: Foundational rules and agent dispatch logic.
  2. agents/<agent>.md: Behavioral constraints for the specific agent.

Antigravity (Gemini)

  1. GEMINI.md: Foundational rules and agent dispatch logic.

Specific agent files override global instructions in case of conflict.

Deployment

  1. Codex CLI:

    • Copy codex/AGENTS.md (or codex_zh/AGENTS.md) to your project root or ~/.codex/.
    • Sync config.toml and the agents/ directory to the same location.
  2. Claude Code:

    • Copy claude/CLAUDE.md (or claude_zh/CLAUDE.md) to your project root.
    • Sync the agents/ directory to the same location under ~/.claude/.
  3. Antigravity (Gemini):

    • Copy gemini/GEMINI.md (or gemini_zh/GEMINI.md) to your project root.

Modifying Instructions

  • Global rules: Update AGENTS.md (e.g., changing coding standards).
  • Agent behavior: Update the corresponding file in agents/.
  • New Agents: Add a .toml/.md file to agents/ and register it in the routing table within AGENTS.md.
Ver o original no GitHub

Bom para

Relacionados