convez.ai
AgenteAutomaçãoAvançado
fbaltor

Claude Config - Configuração Pessoal de Agente

fbaltor/claude-config

Sistema modular de skills, agentes e scripts para estender Claude Code

Ver no GitHub
InicialPluginScore · de 0 a 100

Repositório TypeScript que centraliza configuração pessoal para Claude Code, incluindo sub-agentes especializados, skills customizadas (comandos slash), scripts de integração com Linear/GitHub e instruções globais. Funciona como sistema de extensão para automatizar fluxos de desenvolvimento, pesquisa de codebase, planejamento de implementação e sincronização com ferramentas externas.

Instalação

$npx convez-ai fbaltor/claude-config

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 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Automatizar planejamento e execução de implementações com gates de verificação
  • Pesquisar e analisar codebase com sub-agentes especializados em paralelo
  • Sincronizar documentos entre Linear e repositório local
  • Triagiar comentários de PR no GitHub (Major/Minor/False Positive)
  • Analisar impacto de mudanças em toda a base de código

Veredito

Configuração técnica robusta para power users de Claude Code que precisam orquestrar agentes, integrar APIs (Linear, GitHub) e automatizar fluxos complexos de desenvolvimento. Requer conhecimento de TypeScript, estrutura de agentes e variáveis de ambiente, mas oferece controle fino sobre comportamento e capacidades do assistente.

README

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

claude-config

Personal Claude Code configuration: custom skills, sub-agents, scripts, and integrations.

Note: This repo lives at ~/.claude/ and is loaded automatically by Claude Code. The CLAUDE.md file provides global instructions to the agent; this README is for human onboarding only and does not affect agent behavior.


Repository Structure

~/.claude/
├── agents/          # Sub-agent definitions (spawned by skills)
├── skills/          # Custom slash-command skills
├── scripts/         # TypeScript utilities (Linear, GitHub)
├── plugins/         # Installed Claude Code plugins
├── plans/           # Implementation plans (git add -f to track)
├── research/        # Investigation documents (git add -f to track)
├── CLAUDE.md        # Global agent instructions (auto-loaded)
├── settings.json    # Permissions, plugins, and feature flags
└── statusline.sh    # Context/cost display for the Claude Code UI

Skills

Skills are invoked via /skill-name inside Claude Code. Each lives in skills/<name>/SKILL.md.

SkillCommandDescription
Create Plan/create_planInteractive implementation planning with parallel research agents. Saves to ~/plans/.
Implement Plan/implement_planPhase-by-phase execution of an approved plan with human verification gates.
Impact Analysis/impact_analysisFind all files and lines affected by a proposed change (read-only).
Research Codebase/research_codebaseDeep codebase investigation. Spawns parallel sub-agents. Saves to ~/research/.
Linear/linearFetch Linear issues/projects. Auto-detects issue from git branch (e.g. JUMP-123).
Linear Push Doc/linear-push-docSync a local markdown file to a Linear document.
Linear Pull Doc/linear-pull-docPull a Linear document into a local markdown file.
Triage Reviews/triage-reviewsClassify GitHub PR review comments as Major / Minor / False Positive.

Sub-Agents

Defined in agents/*.md. These are spawned by skills to run specialized tasks in parallel.

AgentModelPurpose
codebase-locatorSonnetFind where files, directories, and components live
codebase-analyzerSonnetUnderstand how specific components work
codebase-pattern-finderSonnetFind existing code patterns and usage examples
impact-analyzerOpusExhaustive inventory of files affected by a change
web-search-researcherSonnetResearch questions using web search

All agents operate as documentarians — they describe existing code without suggesting improvements unless asked.

Scripts

TypeScript utilities in scripts/, managed with npm. Run via npx tsx <script>.

  • linear-fetch.ts — Fetch Linear issues and projects via the Linear SDK
  • linear-doc-sync.ts — Bi-directional markdown-to-Linear document sync
  • reviews/ — GitHub PR review fetching, status checking, and YAML output

Setup

cd ~/.claude/scripts
npm install

Environment Variables

The scripts require API keys set as environment variables. Copy the example file and fill in your keys:

cp ~/.claude/scripts/.env.example ~/.claude/scripts/.env
VariableUsed ByPurposeRequired?
LINEAR_API_KEYLinear scriptsGeneral Linear API authenticationYes, if specific keys below are not set
LINEAR_API_KEY_READLinear scriptsRead-only Linear API access (fetch issues, pull docs)No — falls back to LINEAR_API_KEY
LINEAR_API_KEY_ALLLinear scriptsRead+write Linear API access (push docs)No — falls back to LINEAR_API_KEY
GITHUB_TOKENReview scriptsGitHub API authentication (Octokit)No — falls back to gh auth token

Note: The Linear and Notion MCP integrations (used by Claude Code directly) authenticate via OAuth through the plugin system — they do not need environment variables.

Integrations (MCP)

Configured in settings.json via Model Context Protocol servers:

  • Linear — Full read/write (no delete)
  • Notion — Read/write/create/update (no delete)

Plans and Research

These directories are git-ignored by default to avoid noise. To track a specific file:

git add -f plans/2026-03-20-my-plan.md
git add -f research/2026-03-20-my-investigation.md

Naming convention: YYYY-MM-DD-kebab-case-description.md

Settings Overview

settings.json controls:

  • Permissions — Which tools Claude Code can use without prompting
  • Plugins — Currently: code-simplifier
  • Features — Extended thinking enabled, high effort level, auto-memory on
  • Status line — Custom bash script showing context usage %, cost, and model

settings.local.json holds machine-specific overrides (not intended for sharing).

Workflow Examples

Planning and Implementing a Feature

Use /create_plan and /implement_plan together for structured, multi-phase work.

# 1. Create a plan — Claude researches the codebase and iterates with you
/create_plan Add cursor-based pagination to all listing API endpoints

# Claude will:
# - Spawn sub-agents to find relevant files and patterns
# - Ask clarifying questions
# - Present design options
# - Write a detailed plan to ~/.claude/plans/2026-03-20-add-pagination.md

# 2. Review the plan, suggest changes, iterate until satisfied

# 3. Implement it phase by phase
/implement_plan ~/.claude/plans/2026-03-20-add-pagination.md

# Claude will:
# - Implement Phase 1, run automated verification
# - Pause for you to manually test
# - Continue to Phase 2 after your confirmation
# - Update checkboxes in the plan file as it goes

Triaging PR Reviews

Use /triage-reviews to classify review comments and focus on what matters.

# Triage reviews on a specific PR
/triage-reviews --pr 456 --repo org/repo

# Auto-detect PR from current branch (defaults to Jumpstart-Immigration/jumpstart)
/triage-reviews

# Wait for bot reviews (CodeRabbit, Copilot) to finish before triaging
/triage-reviews --wait --pr 456

# Claude will:
# - Fetch all review comments (human + bot)
# - Read the actual source code at each commented location
# - Classify each as Major / Minor / False Positive
# - Present outdated comments separately
# - Ask which items you want to tackle

Output looks like:

## PR Review Triage

**PR #456**: Add user profile endpoint
**Reviewers**: @alice, @coderabbitai[bot]
**Stats**: 2 major | 3 minor | 1 false positive | 2 outdated

### Major (2)
1. **Missing null check on user lookup** — @alice
   `src/handlers/profile.ts:42` · [link](...)
   > What happens if the user ID doesn't exist?

   **Assessment**: Confirmed — no guard before accessing user properties.
...

Syncing Documents with Linear

Use /linear-push-doc and /linear-pull-doc for one-way syncs between local markdown and Linear documents.

# Pull a Linear document locally for the first time (needs the document ID)
/linear-pull-doc docs/architecture.md --id abc123-def456

# This creates docs/architecture.md with frontmatter:
# ---
# linear_document_id: abc123-def456
# linear_document_title: Architecture Overview
# linear_sync_hash: 5a8f5572fdf9
# ---

# After editing locally, push changes back to Linear
/linear-push-doc docs/architecture.md

# Push ALL Linear-linked docs in the repo at once
/linear-push-doc

# Pull latest changes from Linear into the local file
/linear-pull-doc docs/architecture.md

The document ID is stored in YAML frontmatter after the initial pull, so subsequent syncs don't need --id. A linear_sync_hash is also written on each push/pull — the pre-PR hook uses it to verify docs are synced before creating a PR (only docs changed on the branch are checked).

Fetching Linear Issues

Use /linear to pull issue or project context into your Claude Code session.

# Auto-detect issue from current git branch (parses JUMP-28, GOJ-12, etc.)
/linear --fetch-issue

# Fetch a specific issue
/linear --fetch-issue JUMP-42

# Fetch a project overview with all its issues and docs
/linear --fetch-project "API Redesign"

Investigating a Codebase

Use /research_codebase for deep, documented investigations.

# Research how authentication works in the project
/research_codebase How does the auth middleware handle token refresh?

# Claude will:
# - Spawn parallel sub-agents (locator, analyzer, pattern-finder)
# - Synthesize findings into a structured document
# - Save to ~/.claude/research/2026-03-20-auth-token-refresh.md
# - Present findings with file:line references

Assessing Impact of a Change

Use /impact_analysis before refactoring to know exactly what will break.

/impact_analysis Rename the UserService class to ProfileService

# Claude will:
# - Find every file that imports, references, or tests UserService
# - Classify as: Must change / Should check / Worth reviewing
# - Return a read-only inventory — never modifies files

How to Use This Repo

If you want to adopt parts of this configuration:

  1. Fork this repo
  2. Clone into your own ~/.claude/ directory (back up your existing config first)
  3. Run cd ~/.claude/scripts && npm install for script dependencies
  4. Set up Linear API keys if using Linear integration
  5. Customize CLAUDE.md and settings.json for your own preferences

Important: CLAUDE.md is auto-loaded into every Claude Code conversation as system instructions. Keep it focused and concise — bloated instructions consume context and degrade agent performance.

Claude Code Documentation References

This repo uses several Claude Code extension features. Here are the official docs for each:

FeatureUsed InDocs
Features Overview—Extend Claude Code — When to use skills vs agents vs hooks vs MCP
Skillsskills/Custom slash commands — SKILL.md files, frontmatter, $ARGUMENTS
Sub-agentsagents/Custom subagents — Agent definitions, tools, model selection
Memory & CLAUDE.mdCLAUDE.mdMemory system — Global/project instructions, auto-memory, MEMORY.md
Settingssettings.jsonConfiguration — Permissions, features, scopes
MCP Serverssettings.jsonMCP integrations — Connecting to Linear, Notion, etc.
Pluginsplugins/Plugin system — Bundling skills, agents, and hooks
Status Linestatusline.shCustom status bar — Context usage, cost, model display
Hookssettings.jsonHooks guide — Pre/post tool execution automation
Ver o original no GitHub

Bom para

Relacionados