claude-config
sseely/claude-config
Claude Code agent definitions, rules, skills, and settings
Instalação
npx convez-ai sseely/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
README
A tradução para o português deste README ainda está na fila.
claude-config
Personal Claude Code configuration: agents, rules, skills, and settings.
Credits
The agents in this repo are derived from
VoltAgent/awesome-claude-code-subagents.
Modifications include: stripped boilerplate (fake JSON communication protocols,
workflow phases, MCP tool sections), corrected tool lists to real Claude Code
tools only, added model: assignments (opusplan / sonnet / haiku),
added disallowedTools: on read-only agents, and added memory: user on
select agents.
Several skills (changelog-generator, video-downloader, file-organizer,
internal-comms, webapp-testing) are adapted from
ComposioHQ/awesome-claude-skills.
Modifications include: fixed macOS compatibility issues, corrected skill name
fields to match directory names, trimmed preamble and padding sections, removed
vendor-specific infrastructure requirements, and added missing prerequisites.
The document skills (doc-pdf, doc-docx, doc-pptx, doc-xlsx) are also
sourced from that repo but are excluded from this public repo — they carry a
proprietary Anthropic license (© 2025 Anthropic, PBC) that prohibits
redistribution.
Installation
New machine
# Prerequisites: Claude Code installed, gh authenticated
git clone https://github.com/sseely/claude-config.git ~/.claude
Claude Code reads ~/.claude/ directly — no symlinks or additional setup needed.
Existing ~/.claude/
If you already have a ~/.claude/ directory with content you want to keep,
merge selectively:
git clone https://github.com/sseely/claude-config.git /tmp/claude-config
cp -r /tmp/claude-config/agents ~/.claude/
cp -r /tmp/claude-config/rules ~/.claude/
cp -r /tmp/claude-config/skills ~/.claude/
cp -r /tmp/claude-config/hooks ~/.claude/
cp -r /tmp/claude-config/templates ~/.claude/
cp /tmp/claude-config/CLAUDE.md ~/.claude/CLAUDE.md
Post-install
Two files are excluded from the repo and must be created manually:
~/.claude/settings.json — machine-local settings (permissions, hooks).
See templates/ for examples.
~/.claude/.mcp.json — MCP server config (paths are machine-specific).
Auto-install tools
The session-start hook checks for sg (ast-grep) on every start.
To enable automatic installation:
export CLAUDE_AUTO_INSTALL_TOOLS=true
Without this flag, missing tools are reported but not installed.
What's included
CLAUDE.md
Global instructions loaded at the start of every Claude Code session. Covers interaction style, when to use agents, commit conventions, and pointers to the rules files. Kept under 200 lines (the load limit).
rules/
Instruction files loaded every session alongside CLAUDE.md.
| File | Content |
|---|---|
autonomous-execution.md | Mission-brief protocol for unattended multi-hour sessions — startup sequence, batch execution, quality gates, decision-making rules, compaction recovery |
code-principles.md | YAGNI, SOLID, no magic literals, prefer native fetch over HTTP libraries |
commits.md | Conventional Commits spec — subject format, body criteria, BREAKING CHANGE footer, examples |
parallelism.md | Multi-agent planning rules — file ownership (one writer per file), agent prompt structure, when to parallelize vs. serialize |
security.md | Input validation, secrets management, error response hygiene, auth/authz checklist, injection prevention |
testing.md | TDD (red-green-refactor), 90/90/90 coverage floor, shared test helpers in test/helpers/ |
skills/
Slash-command workflows invoked as /skillname.
| Skill | Usage | Description |
|---|---|---|
code-review | /code-review [scope] | Parallel 10-agent review (correctness, security, linting, error handling, deps, tests, logging, types, perf, API contracts). Defaults to staged changes; pass "full project", a path, or a glob |
commit | /commit | Stage files, draft Conventional Commits message, commit |
explore | /explore | Map an unfamiliar codebase — clones related repos from the same GitHub org, generates architecture diagrams and component maps in architecture-temp/ |
fix | /fix [error] | Drive a failing test or build error to green. Loops debugger → language agent → test runner (max 5 iterations). Accepts a test name, error message, or file:line |
plan-mission | /plan-mission [feature] | Turn a feature description into a mission brief for 1-4 hours of autonomous execution — blast radius analysis, architecture decisions, batched task specs with Given/When/Then acceptance criteria |
review-pr | /review-pr [PR] | Fetch a GitHub PR diff, run the full /code-review checklist, then post findings as inline review comments via the GitHub API. Accepts a PR URL, bare number, or infers from the current branch. Never auto-approves — the human submits |
sandbox | /sandbox [name] [repo] [prompt] | Run a task or mission brief inside an isolated Docker container. Auto-detects languages, assembles a multi-layer Dockerfile, injects secrets from macOS Keychain at runtime, and uses named volumes for resumability |
upgrade-deps | /upgrade-deps [scope] | Parallel dependency audit + security scan across all detected languages. Simple upgrades execute directly; complex ones (breaking changes, multi-language) route to /plan-mission |
hooks/
Shell scripts triggered by Claude Code lifecycle events.
| Hook | Event | Description |
|---|---|---|
record-turn-start.sh | UserPromptSubmit | Writes a Unix timestamp to ~/.claude/.runtime/claude-turn-start |
notify-on-stop.sh | Stop | macOS notification with elapsed time if the turn took >30 seconds |
quality-gate.sh | Manual | Runs project-specific quality checks — reads .claude-quality-gates or auto-detects (Node, Python, Go, Rust, .NET) |
autonomous-toggle.sh | Manual | Copies autonomous permissions into a project's .claude/settings.json (with backup/restore) for unattended sessions |
templates/
| File | Description |
|---|---|
autonomous-settings.json | Broad permission set for autonomous execution — copied into projects by autonomous-toggle.sh |
agents/
128 specialist agents across 10 categories. Invoked automatically by Claude
Code when a task matches their domain, or explicitly via the Agent tool.
Model assignments:
opusplan— architecture agents (complex reasoning, plan-then-execute)sonnet— implementation agents (default workhorse)haiku— review/audit/research agents (read-only, fast, cheap)
01-core-development (10 agents)
api-designer, backend-developer, electron-pro, frontend-developer,
fullstack-developer, graphql-architect*, microservices-architect*,
mobile-developer, ui-designer, websocket-engineer
02-language-specialists (29 agents)
angular-architect, cpp-pro, csharp-developer, django-developer,
dotnet-core-expert, dotnet-framework-4.8-expert, elixir-expert,
flutter-expert, golang-pro, java-architect*, javascript-pro,
kotlin-specialist, laravel-specialist, nextjs-developer, php-pro,
powershell-5.1-expert, powershell-7-expert, python-pro, rails-expert,
react-specialist, ruby-2-7-specialist, ruby-specialist, rust-engineer,
spring-boot-engineer, sql-pro, swift-expert, typescript-pro,
vue-expert, wordpress-master
03-infrastructure (16 agents)
azure-infra-engineer, cloud-architect*, database-administrator,
deployment-engineer, devops-engineer, devops-incident-responder,
docker-expert, incident-responder, kubernetes-specialist,
network-engineer, platform-engineer, security-engineer, sre-engineer,
terraform-engineer, terragrunt-expert, windows-infra-admin
04-quality-security (14 agents)
accessibility-tester, ad-security-reviewer, architect-reviewer,
chaos-engineer, code-reviewer, compliance-auditor, debugger,
error-detective, penetration-tester, performance-engineer,
powershell-security-hardening, qa-expert, security-auditor,
test-automator
05-data-ai (13 agents)
ai-engineer, data-analyst, data-engineer, data-scientist,
database-optimizer, llm-architect*, machine-learning-engineer,
ml-engineer, mlops-engineer, nlp-engineer, postgres-pro,
prompt-engineer
06-developer-experience (13 agents)
build-engineer, cli-developer, dependency-manager,
documentation-engineer, dx-optimizer, git-workflow-manager,
legacy-modernizer, mcp-developer, powershell-module-architect,
powershell-ui-architect, refactoring-specialist, slack-expert,
tooling-engineer
07-specialized-domains (12 agents)
api-documenter, blockchain-developer, embedded-systems,
fintech-engineer, game-developer, iot-engineer, m365-admin,
mobile-app-developer, payment-integration, quant-analyst,
risk-manager, seo-specialist
08-business-product (11 agents)
business-analyst, content-marketer, customer-success-manager,
legal-advisor, product-manager, project-manager, sales-engineer,
scrum-master, technical-writer, ux-researcher
09-meta-orchestration (3 agents)
agent-installer, it-ops-orchestrator, memory-curator
10-research-analysis (7 agents)
competitive-analyst, data-researcher, market-researcher,
research-analyst, scientific-literature-researcher, search-specialist,
trend-analyst
* opusplan model
Dependencies
The following agents are sourced from external repos and should be updated when those repos change:
| Agent | Source repo |
|---|---|
agents/09-meta-orchestration/memory-curator.md | sseely/claude-memory |
Updating
cd ~/.claude && git pull