convez.ai
AgenteAutomaçãoAvançado
lutsenko-yuriy

Yuriys Agentic Boyz

lutsenko-yuriy/yuriys-agentic-boyz

Multi-agente agentic com Claude para ciclo completo de desenvolvimento

Ver no GitHub
InicialPluginScore · de 0 a 100

Template de repositório que bootstrapa um workflow multi-agente com Claude, integrando Product Owner, Tech Lead, Developer e Code Reviewer. Conecta-se a ferramentas de gestão de projetos (Linear, Jira, GitHub Issues) e hosts Git, orquestrando um pipeline completo de análise, design, implementação e revisão de código.

Instalação

$npx convez-ai lutsenko-yuriy/yuriys-agentic-boyz

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 3 semanas
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Automatizar fluxo de desenvolvimento end-to-end com múltiplos papéis de IA
  • Gerar planos de implementação e análises de produto de forma estruturada
  • Executar code review arquitetural e testes em paralelo com IA
  • Sincronizar backlog entre PM tool e git, mantendo documentação atualizada

Veredito

Projeto ambicioso que modela um workflow realista de times de desenvolvimento via multi-agent. Exige setup manual significativo e conhecimento profundo de CI/CD, mas oferece padrão robusto para times que querem escalar coordenação de IA. Recomendado para organizações com processos maduros.

README

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

Multi-Agent Project Template

A GitHub template repository that bootstraps a multi-skill AI workflow with your choice of project management tool (Linear, Jira, GitHub Issues, etc.) and Git host (GitHub, GitLab, Bitbucket) for any new project in minutes.

What's included

FilePurpose
skills/configure/calibrate/SKILL.mdOne-time setup: propose and approve the model → tier mapping
skills/configure/style/SKILL.mdSwitch communication style: DETAILED, CONCISE, or SCHEMATIC
skills/manage/summarize/SKILL.mdSession-start: fetch and display the backlog
skills/manage/ship/SKILL.mdPost-merge housekeeping: close issues, update docs, bump version, merge
skills/design/analyze/SKILL.mdAnalytics planning: identify events and screen views for a feature
skills/design/experiment/SKILL.mdExperiment design: hypothesis, metrics, variant spec, registry entry
skills/design/plan/SKILL.mdImplementation planning: structured plan from a PM issue
skills/build/implement/SKILL.mdTDD implementation and PR/MR
skills/verify/review/SKILL.mdArchitectural PR/MR review
skills/verify/audit/SKILL.mdRuntime and migration PR/MR review
styles/DETAILED.mdStyle def: full prose, default
styles/CONCISE.mdStyle def: lecture-note shorthand, abbreviation-friendly
styles/SCHEMATIC.mdStyle def: TeX-math + Haskell notation
CLAUDE.mdOrchestrator file: session start, workflow steps, branch naming, review chain
docs/PRODUCT_SPEC.mdBlank product spec template
docs/ARCHITECTURE.mdBlank architecture doc template
docs/BACKLOG.mdGenerated from your PM tool — do not edit by hand
docs/CHANGELOG.mdMaintained by the ship skill after each merged PR/MR
docs/VERSIONING.mdVersioning strategy template
docs/MODEL_TIERS.mdEffort Tier + Reasoning Depth vocabulary; active model → tier mapping
setup.shInteractive setup script — substitutes {{PLACEHOLDERS}} throughout

Skill workflow

Session start
  └─▶ summarize — presents backlog, asks "what goes into the next release?"

User-facing feature
  └─▶ analyze — plans analytics events and screen views, waits for approval
        └─▶ plan — produces implementation plan, waits for approval
               └─▶ implement — TDD, opens PR/MR
                      └─▶ review + audit — parallel review (architectural + runtime)
                             └─▶ ship — closes issues, regenerates docs, merges PR/MR

Quick start

1. Create a new repo from this template

Click "Use this template" → "Create a new repository" on GitHub.

2. Clone your new repo

git clone https://github.com/<you>/<your-project>.git
cd <your-project>

3. Run the setup script

chmod +x setup.sh
./setup.sh

The script will prompt for:

  • Project name and description
  • Tech stack and architecture summary
  • PM tool name (Linear, Jira, GitHub Issues, etc.) and issue prefix
  • Git host (GitHub, GitLab, Bitbucket)
  • Available AI models (comma-separated — used by calibrate)
  • Experiment tracking tool
  • AI commit trailer and PR/MR credit lines

All {{PLACEHOLDER}} values across every file are substituted in one step.

4. Finish manual setup

After setup.sh:

  1. CLAUDE.md — Common Commands: fill in the test, lint, build, and install commands for your stack.
  2. docs/PRODUCT_SPEC.md: describe your features from the user's perspective.
  3. docs/ARCHITECTURE.md: add your directory tree and layer rules.
  4. CLAUDE.local.md (not committed): add local binary paths, PM tool auth notes, and any machine-specific config.

5. Map your models to tiers

Invoke the calibrate skill:

Invoke the calibrate skill

This skill uses THOROUGH + ARCHITECTURAL reasoning (your most capable model) to read the model list you provided, characterise each model's strengths, and propose which model should handle each Effort Tier + Reasoning Depth combination. Review and approve the proposal — it gets written into docs/MODEL_TIERS.md as the Active mapping.

Re-run calibrate any time your available models change.

6. Authenticate your PM tool MCP (if applicable)

If your PM tool has an MCP server (e.g. Linear), open Claude Code in your project directory and run /mcp to authenticate. See CLAUDE.local.md for notes.

7. Start your first session

The CLAUDE.md session-start instructions will automatically invoke the summarize skill, which will present an empty backlog and ask: "What goes into the first release?"

Placeholders reference

PlaceholderExample value
{{PROJECT_NAME}}My App
{{PROJECT_DESCRIPTION}}A task manager for remote teams
{{STACK}}Flutter/Dart
{{ARCHITECTURE_SUMMARY}}Vertical-slice with Riverpod + sqflite
{{CODE_STYLE}}Flutter style guide
{{PM_TOOL}}Linear
{{ISSUE_PREFIX}}APP
{{PM_PROJECT_URL}}https://linear.app/my-workspace/project/...
{{GIT_HOST}}GitHub
{{EXPERIMENT_TOOL}}Firebase A/B Testing
{{AVAILABLE_MODELS}}gpt-4o, gpt-4o-mini
{{AI_COMMIT_TRAILER}}Co-Authored-By: AI Assistant <ai@example.com>
{{AI_TOOL_CREDIT}}🤖 Generated with Claude Code

Notes

  • .claude/skills/ is committed — skills are part of the project workflow. If you want them local-only, add .claude/skills/ to .gitignore after setup.
  • No external PM tool required — leave the PM tool blank during setup.sh and the template defaults to your Git host's built-in issue tracker (GitHub Issues, GitLab Issues, etc.). No MCP, no extra auth, no configuration. Add an external tool later if the project outgrows it.
  • PM tool auth is per-developer — when using an external PM tool with an MCP server (e.g. Linear), each team member authenticates independently. No secrets are stored in the repo.
  • CLAUDE.local.md is gitignored — put machine-specific paths and personal notes there.
  • Tool-agnostic by design — skills describe what to do with multi-tool example tables (Linear/Jira/GitHub Issues/GitLab Issues, GitHub/GitLab/Bitbucket). Specific commands depend on your configured tools.
  • Model-agnostic by design — skills declare effort and reasoning tiers instead of model names. The calibrate skill maps your available models to those tiers once, and the active mapping lives in docs/MODEL_TIERS.md.
  • Communication styles — the style skill switches between DETAILED (full prose), CONCISE (lecture-note shorthand), and SCHEMATIC (TeX-like notation). Active style persists across sessions via CLAUDE.local.md.
Ver o original no GitHub

Bom para

Relacionados