Claude Config - Configuração Pessoal de Agente
fbaltor/claude-config
Sistema modular de skills, agentes e scripts para estender Claude Code
Ver no GitHubRepositório 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-config1 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
- 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. TheCLAUDE.mdfile 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.
| Skill | Command | Description |
|---|---|---|
| Create Plan | /create_plan | Interactive implementation planning with parallel research agents. Saves to ~/plans/. |
| Implement Plan | /implement_plan | Phase-by-phase execution of an approved plan with human verification gates. |
| Impact Analysis | /impact_analysis | Find all files and lines affected by a proposed change (read-only). |
| Research Codebase | /research_codebase | Deep codebase investigation. Spawns parallel sub-agents. Saves to ~/research/. |
| Linear | /linear | Fetch Linear issues/projects. Auto-detects issue from git branch (e.g. JUMP-123). |
| Linear Push Doc | /linear-push-doc | Sync a local markdown file to a Linear document. |
| Linear Pull Doc | /linear-pull-doc | Pull a Linear document into a local markdown file. |
| Triage Reviews | /triage-reviews | Classify 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.
| Agent | Model | Purpose |
|---|---|---|
| codebase-locator | Sonnet | Find where files, directories, and components live |
| codebase-analyzer | Sonnet | Understand how specific components work |
| codebase-pattern-finder | Sonnet | Find existing code patterns and usage examples |
| impact-analyzer | Opus | Exhaustive inventory of files affected by a change |
| web-search-researcher | Sonnet | Research 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 SDKlinear-doc-sync.ts— Bi-directional markdown-to-Linear document syncreviews/— 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
| Variable | Used By | Purpose | Required? |
|---|---|---|---|
LINEAR_API_KEY | Linear scripts | General Linear API authentication | Yes, if specific keys below are not set |
LINEAR_API_KEY_READ | Linear scripts | Read-only Linear API access (fetch issues, pull docs) | No — falls back to LINEAR_API_KEY |
LINEAR_API_KEY_ALL | Linear scripts | Read+write Linear API access (push docs) | No — falls back to LINEAR_API_KEY |
GITHUB_TOKEN | Review scripts | GitHub 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:
- Fork this repo
- Clone into your own
~/.claude/directory (back up your existing config first) - Run
cd ~/.claude/scripts && npm installfor script dependencies - Set up Linear API keys if using Linear integration
- Customize
CLAUDE.mdandsettings.jsonfor your own preferences
Important:
CLAUDE.mdis 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:
| Feature | Used In | Docs |
|---|---|---|
| Features Overview | — | Extend Claude Code — When to use skills vs agents vs hooks vs MCP |
| Skills | skills/ | Custom slash commands — SKILL.md files, frontmatter, $ARGUMENTS |
| Sub-agents | agents/ | Custom subagents — Agent definitions, tools, model selection |
| Memory & CLAUDE.md | CLAUDE.md | Memory system — Global/project instructions, auto-memory, MEMORY.md |
| Settings | settings.json | Configuration — Permissions, features, scopes |
| MCP Servers | settings.json | MCP integrations — Connecting to Linear, Notion, etc. |
| Plugins | plugins/ | Plugin system — Bundling skills, agents, and hooks |
| Status Line | statusline.sh | Custom status bar — Context usage, cost, model display |
| Hooks | settings.json | Hooks guide — Pre/post tool execution automation |