convez.ai
MCP ServerAutomaçãoAvançado
shenxingy

Clade — Sistema Autônomo de Codificação

shenxingy/Clade

Codificação autônoma com segurança, aprendizagem contínua e orquestração inteligente

Ver no GitHub
BomPluginScore · de 0 a 100

MCP server que transforma Claude Code em um sistema autônomo de codificação com 124 skills, 27 hooks e 36 agentes. Inclui guardião de segurança, loop de aprendizagem por correção e execução paralela com gates de qualidade para código robusto sem supervisão.

Instalação

$npx convez-ai shenxingy/Clade

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
10
Forks
6
Acompanhando
0
Saúde
100%
Último commit
há 2 dias
Versão
v0.2.0
Issues abertas
2
Contribuidores
3

Casos de uso

  • Automação de tarefas de codificação complexas executadas sem supervisão
  • Detecção e correção automática de erros em código (type-checking, padrões recorrentes)
  • Integração em Cursor, Windsurf, Claude Desktop para acesso a 124 skills especializadas
  • Implementação de pipelines CI/CD com verificação de qualidade automática
  • Aprendizagem incremental a partir de correções manuais para melhorar futuros prompts

Veredito

Clade é uma solução sofisticada e bem-arquitetada para quem precisa de codificação autônoma industrial-grade. A combinação de guardião de segurança, hooks contextuais e loop de aprendizagem por correção eleva a confiabilidade, mas requer familiaridade com git, shell scripts e CI/CD. Recomendado para equipes que usam Claude Code regularmente em projetos complexos.

README

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

English | 中文

Clade

Autonomous coding, evolved.

124 skills, 27 hooks, 36 agents, a safety guardian, and a correction learning loop — all working together so Claude codes better, catches its own mistakes, and can run unattended overnight while you sleep.

Two layers: CLI (product center — skills, hooks, scripts you live in) + Orchestrator (observability + gates + execution adapter — watch parallel workers, verify quality gates, route work).

If this saves you time, a star helps others find it. Something broken? Open an issue.

Blog post: Building Clade — motivation, design decisions, and lessons learned.

Table of Contents

  1. Install
  2. MCP Server
  3. What It Does
  4. Self-Learning Mechanisms
  5. Skills
  6. Supported Languages
  7. Documentation
  8. Repo Structure
  9. Contributing
  10. License

Install

Full Framework (recommended)

git clone https://github.com/shenxingy/clade.git
cd clade && ./install.sh

Installs skills, hooks, agents, scripts, and safety guardian. Start a new Claude Code session to activate.

Requires: jq. Platform: Linux and macOS.

MCP Server Only

If you just want the skills in Cursor, Windsurf, Claude Desktop, or any MCP client:

pip install clade-mcp

See MCP Server below for configuration.

MCP Server — Use Skills in Any AI Editor

The MCP server exposes all 124 Clade skills as callable tools via the Model Context Protocol. Works with any MCP-compatible client.

Claude Desktop / Claude Code:

{
  "mcpServers": {
    "clade": { "command": "uvx", "args": ["clade-mcp"] }
  }
}

Cursor / Windsurf:

{
  "mcpServers": {
    "clade": { "command": "clade-mcp" }
  }
}

Prerequisite: Claude Code CLI must be installed — skills execute via claude -p.

What It Does

WhenWhat firesEffect
Session opens in a git reposession-context.shLoads git context, handoff state, correction rules, model guidance
Session opens in a git repocommit-archeology.shMines git log for recurring fix patterns (wiring/deploy/compat gaps, Claude-overridden) — injects top 4
Claude runs a bash commandpre-tool-guardian.shBlocks dangerous ops: migrations, rm -rf, force push, DROP TABLE
Claude edits codepost-edit-check.shAsync type-check (tsc, pyright, cargo check, go vet, etc.)
You correct Claudecorrection-detector.shLogs correction, prompts Claude to save a reusable rule
Claude marks task doneverify-task-completed.shAdaptive quality gate: compile + lint, build + test in strict mode

See How It Works for the full hook reference (27 hooks).

Self-Learning Mechanisms

Two mechanisms keep Clade aligned with reality without manual upkeep:

  • Commit Lessons (reactive) — commit-archeology.sh mines git log for recurring fix patterns (wiring-gap, deploy-gap, compat-gap, claude-overridden) and injects the top 4 at every session start.
  • Doc Align (preventive) — doc-align.py declares shared facts in docs/facts.json (auto-derived from filesystem); checks/auto-fixes drift across every *.md. A PostToolUse hook flags drift the moment you edit a doc, so stale counts never reach commit.

Both work on any project Claude Code is run in (universal, in ~/.claude/scripts/). Both are silent no-ops on repos that haven't opted in.

See Self-Learning Mechanisms for full details, detectors, schemas, and tunable env vars.

Skills (124)

Core Workflow

SkillWhat it does
/commitSplit changes into logical commits by module, push by default
/syncCheck off completed TODOs, append session summary to PROGRESS.md
/review8-phase coverage review — finds AND fixes issues, loops until clean
/verifyVerify project behavior anchors (compile, test, lint)

Autonomous Operation

SkillWhat it does
/startAutonomous session launcher — morning brief, overnight runs, cross-project patrol
/loop GOALGoal-driven improvement loop — supervisor plans, workers execute in parallel
/iloop TASKIn-session iterative loop — Stop hook re-prompts until done (no background workers)
/batch-tasksExecute TODO steps via unattended sessions (serial or parallel)
/orchestrateDecompose goals into tasks for worker execution
/handoffSave session state for context relay between agents
/pickupResume from previous handoff — zero-friction restart
/worktreeCreate git worktrees for parallel sessions
/pokeHeartbeat after esc — 3-line status, auto-continues if still progressing
/statusSession dashboard — background agents, loops, worktrees, unpushed commits
/goExecute the recommendation from your most recent A/B/C option set

Code Quality

SkillWhat it does
/review-pr NAI code review on a PR diff — Critical / Warning / Suggestion
/merge-pr NSquash-merge PR and clean up branch
/investigateRoot cause analysis — no fix without confirmed hypothesis
/incident DESCIncident response — diagnose, postmortem, follow-up tasks
/csoSecurity audit (OWASP + STRIDE)
/mapGenerate ARCHITECTURE.md with module graph + file ownership

Research & Planning

SkillWhat it does
/research TOPICDeep web research, synthesize to docs/research/
/model-researchLatest Claude model data + auto-update configs
/next"What's next?" — fast 1-shot recommendation (default); /next deep for multi-round interview
/briefMorning briefing — overnight commits, costs, next steps
/retroEngineering retrospective from git history
/frontend-designCreate production-grade frontend interfaces

System

SkillWhat it does
/auditClean up correction rules — promote, deduplicate, remove stale
/document-releasePost-ship doc sync (README, CHANGELOG, CLAUDE.md)
/pipelineHealth check for background pipelines
/providerSwitch LLM provider
sltToggle statusline quota pace indicator

Blog & Content (30 skills)

SkillWhat it does
/blogFull lifecycle — brief → outline → write → SEO check
/blog-writeWrite SERP-informed articles from scratch
/blog-rewriteOptimize existing posts for quality and SEO
/blog-auditFull-site health scan (thin content, meta, cannibalization)
+ 26 moreanalyze · audio · brand · brief · calendar · cannibalization · chart · cluster · discourse · factcheck · flow · geo · google · image · locale-audit · localize · multilingual · notebooklm · outline · persona · repurpose · schema · seo-check · strategy · taxonomy · translate

SEO (25 skills)

SkillWhat it does
/seoFull SEO audit suite
/seo-technicalCrawlability, indexability, Core Web Vitals
/seo-pageDeep single-page analysis
/seo-contentE-E-A-T and content quality scoring
+ 21 moreaudit · backlinks · cluster · competitor-pages · content-brief · dataforseo · drift · ecommerce · flow · geo · google · hreflang · image-gen · images · local · maps · plan · programmatic · schema · sitemap · sxo

Paid Ads (23 skills)

SkillWhat it does
/adsMulti-platform ads audit suite
/ads-googleGoogle Ads — Quality Score, PMax, bidding
/ads-metaMeta Ads — Pixel/CAPI, creative fatigue, Advantage+
/ads-createCreate new ad campaigns from brief
+ 19 moreamazon · apple · attribution · audit · budget · competitor · creative · dna · generate · landing · linkedin · math · microsoft · photoshoot · plan · server-side-tracking · test · tiktok · youtube

Email (6 skills)

SkillWhat it does
/email-writeCompose high-converting emails (PAS, AIDA, BAB frameworks)
/email-auditDeliverability audit — SPF, DKIM, DMARC, blacklists, health score
/email-sequenceDesign automation sequences (welcome, nurture, re-engagement)
+ 3 morecheck · plan · review

See When to Use What for detailed usage guidance.

Supported Languages

Auto-detected — hooks and agents adapt to your project:

LanguageEdit checkType checkerTest runner
TypeScript / JavaScripttsc (monorepo-aware)tscjest / vitest
Pythonpyright / mypypyright / mypypytest
Rustcargo checkcargo checkcargo test
Gogo vetgo vetgo test
Swift / iOSswift buildswift buildswift test
Kotlin / Android / Javagradlewgradlewgradle test
LaTeXchktexchktex—

All checks are opt-in by detection — if the tool isn't installed, the hook silently skips.

Documentation

GuideContents
Maximize ThroughputSkip permissions, batch tasks, parallel worktrees, terminal + voice
Orchestrator Web UIChat-to-plan, worker dashboard, settings, iteration loop
Overnight OperationTask queue, parallel sessions, context relay, safety
How It WorksHooks, agents, skills internals, correction learning, model selection
ConfigurationSettings, thresholds, adding custom hooks/agents/skills
When to Use WhatDetailed usage guidance for every skill
Who to Learn FromVetted watch-list of the agentic-coding frontier — people, repos, bot behavior, reviewed quarterly

Dotfile Sync

Keep ~/.claude/ in sync across machines — memory, corrections, skills, hooks, and scripts.

~/.claude/scripts/sync-setup.sh            # auto-detect NFS or GitHub
~/.claude/scripts/sync-setup.sh --github   # explicit GitHub backend

Fully automatic once configured. See Configuration for details.

Architecture

CLI layer (configs/ → installed to ~/.claude/): Product center. Skills, hooks, scripts you invoke daily. Works standalone, no server needed.

Orchestrator (orchestrator/): Observability + gates + execution adapter. Watches parallel workers, enforces quality gates, routes tasks, provides web UI for multi-project oversight. Optional — CLI works independently.

Web UI (orchestrator/web/): Read-only observation window. Task queue, worker status, cost dashboard, settings. No production logic — all executes via CLI.

clade/
├── install.sh               # One-command deployment
├── configs/                 # ← THE PRODUCT CENTER
│   ├── skills/              # 124 skill definitions
│   ├── hooks/               # 27 event hooks
│   ├── agents/              # 36 agent definitions
│   └── scripts/             # 38 shell + Python utilities
├── orchestrator/            # ← THE EXECUTION ADAPTER
│   ├── server.py            # FastAPI app, routes, WebSocket
│   ├── worker.py            # WorkerPool, SwarmManager, task dispatch
│   ├── task_queue.py        # SQLite task queue + CRUD
│   └── web/                 # ← THE OBSERVATION WINDOW
│       └── src/             # React + Vite UI (served from dist/)
├── docs/                    # Guides and research
├── adapters/openclaw/       # OpenClaw integration (mobile monitoring)
└── templates/               # Settings, CLAUDE.md templates

OpenClaw Integration

Monitor and control overnight loops from your phone via OpenClaw.

SkillTriggerEffect
clade-status"how's the loop going"Iteration progress, cost, commits
clade-control"start a loop to fix tests"Start/stop autonomous loops
clade-report"what did it do overnight"Session report, cost breakdown

See adapters/openclaw/README.md for setup.

Contributing

Contributions welcome — code, docs, issue triage, bug reports. See CONTRIBUTING.md.

Known Limitations

  1. Loop on non-code tasks (research/docs) fails silently — workers produce no diff, loop reports failure
  2. Workers inherit parent env — project-specific env vars leak into worker shells; sanitize before overnight runs
  3. Context budget is per-session — multi-day runs may exhaust context; use /handoff + /pickup

License

MIT

Ver o original no GitHub

Bom para

Relacionados